VISUAL PERFORMANCE UNDER CMH AND HPS LIGHTING SYSTEMS: NUMELITE PROJECT FINAL REPORT

Transcription

1 TRL Limited PUBLISHED PROJECT REPORT PPR043 VISUAL PERFORMANCE UNDER CMH AND HPS LIGHTING SYSTEMS: NUMELITE PROJECT FINAL REPORT Version: 1.0 by G I Crabb, R J Beaumont, D P Steele, P Darley and M H Burtwell (TRL Limited) Prepared for: NNE : An Integrated Approach to Designing High Intensity Discharge Lighting Systems Clients: Research Directorate-General: European Union, UK Department for Transport and UK County Surveyors Society, Copyright TRL Limited June 2005 This report has been prepared for Research Directorate-General: European Union, UK Department for Transport and UK County Surveyors Society. The views expressed are those of the authors and not necessarily those of Research Directorate-General: European Union, UK Department for Transport and UK County Surveyors Society. Project Manager Approvals Quality Reviewed

2 This report has been produced by TRL Limited, under/as part of a Contract placed by Research Directorate- General: European Union, UK Department for Transport and UK County Surveyors Society. Any views expressed are not necessarily those of Research Directorate-General: European Union, UK Department for Transport and UK County Surveyors Society. TRL is committed to optimising energy efficiency, reducing waste and promoting recycling and re-use. In support of these environmental goals, this report has been printed on recycled paper, comprising 100% postconsumer waste, manufactured using a TCF (totally chlorine free) process

3 Table of contents EXECUTIVE SUMMARY Page i Abstract 1 Chapter 1 Introduction Literature review Laboratory tests Full-scale trial 4 Chapter 2 The application of mesopic vision models to street lighting Background Objectives Human visual response to low (mesopic) light levels The human visual system The retina Adaptation to different light levels Colour vision Summary Light and measurements Zones of vision The spectral sensitivity of the visual system Measurements under conditions departing from the photopic zone Summary Literature review Mesopic vision models Experiments to determine the effects of altered lamp spectra on vision Effects of reduced luminance on visibility Effects of glare on visibility Road user visibility models The driving task Visual factors affecting the driving task TRL/ICE experiments Design of laboratory and field experiments Instruments Conclusions 33 Chapter 3 Laboratory experiments measuring reflectance on three pavement surfaces using CMH and HPS lamps Objectives Test samples Road surface reflectance measurements in lighting design Experimental design Experimental apparatus Instruments Photometric camera Spectroradiometer 40 TRL Limited PPR043

4 3.6.3 Powermeter Experimental Method Measurement of illuminance Measurement of sample luminance Wet condition Results Road surface illuminance measurements Photometric camera Spectroradiometer Surface dressed small road system surface Sample surface from the Rue Berchère Albi, France Stone mastic asphalt road machine surface Comparison of incident and reflected spectra Discussion of results Conclusions 59 Chapter 4 Assessment of visual performance under CMH and HPS lighting systems in a full-scale trial Description of lighting installation Lamp types CMH and HPS Selection of observers Central vision experiment Selection of vehicle speed Selection of target and location Peripheral vision observations Trial parameters Measurements of SRS trial lighting system components Target area measurements Target design and selection for central vision experiment Peripheral vision experiment Power consumption Experimental procedure Central vision task Peripheral vision experiment Questionnaire Supplementary measurements Results Central vision Potential reasons for differences in detection distances at target positions A and E Peripheral vision Questionnaire Discussion Central vision Peripheral Questionnaire System controls Conclusions 107 Chapter 5 Overall conclusions from the project Literature review Laboratory Tests Track Trials 110 TRL Limited PPR043

5 Chapter 6 Implications for lighting design Dimming Glare Energy saving Power factor under load and in standby Summary of design implications 113 Chapter 7 Acknowledgements 115 Chapter 8 References 115 Appendix A. Details of previous TRL experimental work 119 TRL/ICE experiments to determine the effects of lamp type, glare and age 119 A.1 Main model trials 119 A.1.1 Luminance and observer age 119 A.1.2 Headlamp glare 119 A.1.3 Lamp type 119 A.1.4 Target location 120 A.2 Indoor validation trials 120 A.2.1 Lamp type 120 A.2.2 Road surface luminance 120 A.2.3 Glare 121 A.3 Outdoor validation trials 121 A.4 Discussion 121 Appendix B. British and European standards 123 Appendix C. Lamp data from the literature and NumeLiTe targets 127 Appendix D. Bibliography 129 Appendix E. Glossary 133 Appendix F. Laboratory test tables 135 Appendix G. Technical details 145 G.1 Photometric camera 145 G.2 Spectroradiometric illuminance meter and luminance spotmeter 145 Appendix H. Trial questionnaire 147 Appendix I. Track trial data tables 149 TRL Limited PPR043

6

7 EXECUTIVE SUMMARY Scope of the study Highly efficient light sources are under development to reduce the power consumption of lighting systems and hence greenhouse-gas emissions in line with the international Kyoto agreement. Preliminary calculations suggested that, within Europe, the expected annual energy savings could be around 10 terawatt hours (TWh), which corresponds to a reduction of at least 5 million tons of carbon dioxide (CO 2 ) emissions. Such systems, applied to street lighting, should simultaneously produce benefits in road safety, energy efficiency and the visual quality of outdoor lighting schemes. The European "NumeLiTe" (Numerical Light for Technology) project aimed to prove the feasibility of an optimal and energy efficient outdoor city lighting scheme for urban illumination. The European Commission, the UK Department for Transport (DfT), and the UK County Surveyors Society (CSS) were joint sponsors for the part of this three-year research project carried out at TRL. Representatives of six European countries, France, Germany, Greece, Portugal, Switzerland and the United Kingdom were consortium partners in the project. The project primarily targeted the vision of the human eye at low (mesopic) light levels. Specific objectives of the project were to produce: a 60% reduction in energy consumption (operating costs) a 30% reduction in maintenance costs highly improved light quality 30% increase in luminous efficacy and 40% increase in the colour rendering index a 50% reduction in system average warm-up time Moreover, because of the complexity of the human visual system and the road-user s task, evidence was needed to confirm that reduced illumination will not have any adverse effect on road user behaviour and road safety. In the context of this report, road user refers to both pedestrians and drivers. This is the final report on TRL s work on the NumeLiTe project. The work TRL performed can be divided into three parts. The first was a literature review of international reports and papers, carried out by TRL in order to better understand the human vision system in the mesopic range of the eye. The second was a series of laboratory tests, performed to measure and compare the reflectance of selected road surface samples, under illumination from a prototype ceramic metal halide (CMH) lamp and a reference high pressure sodium (HPS) lamp. The results of this experiment were used to plan the third part of the work, a full-scale street lighting trial carried out at TRL, and a pilot study undertaken by other consortium partners in Albi, France. Summary This is the final report by TRL on the NumeLite project. It describes all of the work performed by TRL on the NumeLiTe project. Chapter 2 of the report covers the results of the literature review of outdoor lighting research and human vision. It describes the human visual system in relation to low (mesopic) light levels and comments on research on road user tasks at these light levels. A number of key conclusions are drawn from the literature review: Published values for the wavelengths of peak sensitivity of the eye for the three types of coloursensitive receptors (cones) vary at least from 440nm to 498nm (blue) for the S-cones (short), 535 to 545nm (green) for the M-cones (medium) and 564nm to 580nm (red) for the L-cones (long). In the conventional model of vision, luminance information is derived from the summation of the TRL Limited i PPR043

8 signals from all three of these channels, while chrominance information is derived from the differences between them. The brain uses these chrominance signals to compute the sensation of colour. Other published work gives evidence of four types of receptor in the human eye, which have peak sensitivities at wavelengths of 342nm, 437nm, 532nm and 625nm (although the shortest wavelength of these is suppressed by absorption in the lens in the human eye). In particular, the conventional model of monochromatic rods and colour cones is questioned. The Commission International d Eclairage (CIE) curves for the overall spectral response of the eye are greatly smoothed versions of the true response, based on inadequate measurement capabilities in the 1930's. In the mesopic zone, lighting that stimulates the short and medium wavelength receptors is likely to be more effective than the longer wavelengths. The shape of the visual response curve changes with the adaptation level throughout the mesopic zone, in a complex and poorly defined manner. As luminance decreases through the mesopic zone, there is a loss of colour sensitivity. Photopic measurements are not a reliable guide to the performance of the visual system particularly at the levels used in street lighting. Photopic luminance does not correlate well with perceived brightness. Visual reaction time is considered to have a more direct influence on the hazard avoidance performance of road users than visual acuity. Detection times or detection distances of objects are considered the most appropriate parameters to the road user s task. Performance of the peripheral field is also important to object detection speed. Brightness matching has not been successful in defining the mesopic vision model and is not necessarily the most important to road users under street lighting. Evidence suggests that visual acuity and target detection speeds are more appropriate measures. Reaction times are not significantly affected by opposing headlamp glare. As their luminance levels increase glare is increased. Older drivers take longer to recover from exposure to glare. Discomfort glare is task dependent and may be more important overall than disability glare, as discomfort glare is more difficult to prevent at street lighting levels. The road user task depends on both central and peripheral vision. Several experiments have shown that central vision is little affected by changes in the spectral power distribution (SPD) of the lighting. However, the peripheral field is very sensitive to the SPD of the lighting, which is important for the detection of hazards. None of the British and European standards attempt to account for spectral effects and all are based on photopic lumen measurements. Chapter 3 describes the laboratory experiments, in which a test rig was constructed and equipped with optical measurement equipment. Tests were carried out on road pavement samples from the road system at TRL, the road network in Albi and a third sample from TRL to represent an additional surfacing material. The TRL road system and Albi samples were tested in both the dry condition and the wet condition. Reflectance parameters (Q 0 and S1) for each surface were estimated by comparison of the measurements with standard reflectance tables published by the CIE. The experimental work carried out confirmed that, for all surfaces and both lamp types, there was no significant difference in the spectra of incident and reflected light. This result provides the basis for further work on the full-scale trial. TRL Limited ii PPR043

9 Chapter 4 describes the full-scale trial in the final phase of TRL s work. This compared observers visual performance under: the standard GE Lucalox high pressure (HPS) lamp chosen as a reference, at full-power (N150W) a prototype GE ceramic metal halide (CMH) lamp at full power (N150W) each lamp type at approximately half its luminous output (N95W) Five lamp columns were installed on the TRL Small Road System. These were erected at a spacing of 30m and a height of 9m to match the geometry of the pilot study in Albi. Each column was equipped with two Thorn Decostreet Europa luminaires side-by-side, so that the light source could be changed from HPS to CMH without the need for physical access. The luminaires in each pair were mounted far enough apart to avoid interference. TRL carried out 27 experimental sessions, each of about four hours duration, during April June 2004 with six observers. The experiment included two types of observation, designed to test separately the responses of central and peripheral vision. The observers were asked, while driving a vehicle, to detect a 200mm diameter spherical target placed on the road in one of six positions. The distances at which the target was detected were then recorded. In the peripheral vision observations the time taken for each observer to detect a change of luminance in a peripheral target was measured. The results showed that, for the luminance conditions assessed, there was no significant difference between peripheral reaction times or target detection distances under the different lighting systems, either at full-power or dimmed. This provides some justification that it may be safe to reduce lighting levels, in streets normally lit to 1.0cd/m 2, to 0.5cd/m 2, at least during the quieter part of the night. Dimming to this lower level, using either type of lamp, should be free of any significant visual degradation in visual performance and allow potential energy savings to be realised. Further research at TRL is required to investigate visual performance at lower levels, in the region of 0.2 cd/m 2, under 70W CMH and HPS lamps in the trial lighting installation. At these lower levels the expected benefit of the CMH light should become more apparent. TRL Limited iii PPR043

10

11 Abstract Outdoor lighting, especially for roadways, is one of the most important areas of power consumption as well as being an important safety issue. Any reduction in the energy used by lighting therefore needs to be controlled without compromising road safety. A three-year research project, NNE , entitled An integrated approach to designing high intensity discharge lighting systems, partsponsored by the European Commission, aimed to prove the feasibility of an optimal and energy efficient outdoor lighting scheme for urban illumination. Six European countries: France, Germany, Greece, Portugal, Switzerland and the United Kingdom were consortium partners in the project. The part of the work carried out at TRL was jointly sponsored by the European Commission, the UK Department for Transport (DfT), and the UK County Surveyors Society (CSS). A primary issue to be addressed was whether whiter light has advantages in visual perception terms compared to the light spectrum, of yellow appearance, from conventional lamps and, if so, can it be produced economically with low energy consumption. The vision of the human eye at low (mesopic) light levels and possible efficiency gains in the lamp, luminaire optics and the control gear have therefore being investigated. The first phase of the work at TRL was an international literature review of vision and outdoor lighting research which comments on the effectiveness of the visual system in road user tasks at these low light levels. The second phase comprised laboratory work carried out by TRL which is also described in this report. This involved assessment of the comparative reflective properties of relevant road surfacing materials under conventional and prototype light sources. The samples which were tested were from the trial road at TRL and one of the roads in the trial in Albi, France. Also included was a sample of a negatively textured thin surfacing, which had been trafficked on the TRL road machine. The measurements were taken in a specially built rig at TRL. Measurements were made on all three of the samples while dry and the TRL and Albi surfaces when wet. The results showed that there was no measurable change in the spectrum of the light from either source on reflection by any of the surfaces and allowed calculation of the specularity factors (S1) and estimates of the average luminance coefficients Q 0 to be made. In the third and final phase of the work at TRL full-scale trials were performed. These included observations designed to measure, separately, the responses of road users central and peripheral vision. The trial made use of three male and three female observers covering a wide age range. One test compared the reaction times of the observers in CMH and HPS light, full and dimmed to approximately half power to a change of luminance of a peripheral visual target. In the other test the observers were asked, while driving a vehicle, to detect a 200mm diameter spherical target placed on the road ahead in one of 6 positions, under the same four lighting conditions. The distances at which the target was detected were then recorded. In the peripheral visual observations, the time taken for each observer to detect a change of luminance in a peripheral target, was measured. The results showed that, for the luminance conditions assessed, there was no significant difference between peripheral reaction times or target detection distances under the different lighting systems, either at full-power or dimmed. TRL Limited 1 PPR043

12

13 1 Introduction Highly efficient light sources are sought to help governments reduce the power consumption of lighting systems (some 15% of the total electric power generated is consumed by lighting) and hence the emission of greenhouse gases. Outdoor lighting, especially for roadways, is one of the most important areas of power consumption as well as being an important safety issue. Any reduction in the energy used by lighting needs however to be controlled without compromising road safety. Lamp improvements, coupled with optimisation of the ballast characteristics and optical properties of the fixture, together with potential gains from sources specifically designed for human visual perception under mesopic lighting conditions (N10-3 to 3 cd/m 2 ) have the potential to reduce energy usage in roadway lighting by at least a factor of two. If this potential is realised then, within Europe, the expected energy savings could be of the order of 10 terawatt hours (TWh), which corresponds to a reduction of at least five million tonnes of carbon dioxide (CO 2 ). In order to realise the potential energy saving it would be necessary for the expected improvements in both light distribution and visual effectiveness to permit either a reduction in the number of lighting units, or a reduction in lamp power. Particularly for upgrading existing schemes, a lower powered lamp than currently used, in an improved luminaire, might be substituted on existing columns. The study described was performed under the European Commission part-funded project NumeLiTe (NNE ), entitled An integrated approach to designing high intensity discharge lighting systems. The European Commission, the UK Department for Transport (DfT), and the UK County Surveyors Society (CSS) were joint sponsors for the part of the three-year research project carried out at TRL. Six European countries, France, Germany, Greece, Portugal, Switzerland and the United Kingdom were consortium partners in the project. The project as a whole examined the possibilities for saving energy in street lighting, with particular reference to the use of broad-spectrum white light sources. Topics covered in depth by the NumeLite consortium include improvement in the efficacy, stability and life of lamps, improvement in the design of luminaire reflectors and enclosures to control the light distribution, development of electronic dimming systems and development of networked remote control and monitoring systems. Finally a demonstration lighting scheme was installed and monitored in Albi in the south of France. The key overall objective of the project was to prove the feasibility of a scheme to reduce the energy consumption and costs of urban street lighting, while maintaining public safety. As well as reducing the consumption of non-renewable energy resources, the project provides evidence on how to contribute to reductions in carbon dioxide required by the Kyoto agreement. Contributions to the key objective were sought from several areas of investigation: An optimised ceramic metal halide light source with high efficacy and good colour rendering. Utilisation of the human vision system s response to shorter wavelength light at low (mesopic) light levels. An optimal luminaire design to maximise the use of the available light and minimise glare. The use of low-loss electronic control gear combined with dimming to minimise energy consumption and match light output to conditions. The use of a digital network to monitor and control installations. A further target is enhancement of road safety by providing better visual comfort and good colour perception of traffic signs. At the start of the project it was expected that lamps would meet these aims with a luminous efficacy of at least lumen per watt producing light with a colour rendering index (CRI) of (See Appendix E - Glossary). Current lighting schemes based on 150W high-pressure sodium lamps present an efficacy of 100 lumen per watt with a CRI of TRL Limited 3 PPR043

14 1.1 Literature review This report includes the results of an international literature review of outdoor lighting research describing the human visual system in relation to low (mesopic) light levels, road user tasks. There is evidence in the literature that peripheral visual reaction times are shorter under white light. It is therefore possible that it might be safe to use lower lighting levels, and thus less energy, if this type of light is used. The benchmark for the comparison was a standard 150W high-pressure sodium lamp (HPS also designated SON by lamp manufacturers) with a ferromagnetic ballast and without dimming. This was chosen because it was seen as that most commonly used in urban lighting schemes. A prototype CMH lamp of 150W is the subject of the project but development of 250W and 400W CMH lamps to give similar characteristics are also envisaged. It was hoped the prototype lighting had the potential to reduce the level of illumination of urban and residential lighting schemes without any detrimental effects on road safety and road user (pedestrian and driver) behaviour and comfort. 1.2 Laboratory tests In order to investigate this, TRL first performed laboratory tests comparing the reflectance of selected road surface samples, when illuminated by a prototype ceramic metal halide (CMH) lamp and when illuminated by a reference high pressure sodium (HPS) lamp. The measurements were needed in order to detect any differences and to calculate the expected luminance distribution in the full-scale street lighting trial that was carried out at TRL and also in the pilot study carried out in Albi, France in The NumeLiTe project aimed to utilise possible benefits, to human vision at low-light levels, of the shorter wavelengths present in CMH light. If lower light levels than currently specified can be used this will contribute to the overall aim of reducing energy consumption. The laboratory experiments were designed to show any changes in the spectrum of the reflected light compared with the spectrum of the light incident on the surfaces and to allow calculation of the specularity factors (S1) and estimates of the average luminance coefficients Q 0 to be made. 1.3 Full-scale trial Using the information gained from the laboratory tests a full-scale trial was performed to compare road user performance under lighting from the prototype CMH (broad spectrum white light) and conventional HPS (narrower spectrum light of orange appearance) lamps. The study used two lighting levels with each type of lamp, and six observers representing a cross-section of the population in terms of age and sex. Two types of measurement were made. The first was to measure, while driving, any differences in observers detection of a small target on the road surface. The second was to measure, from a stationary, vehicle their reaction times to a change in a target placed in their peripheral field of vision,. Details of the trial design are provided in Chapter 4 with lamp and target specific measurements described in Section 4.6. Section 4.7 sets out the specific details of the central and peripheral vision experiments respectively. The results and analysis are given in Sections 4.8 and 4.9. Many technical terms used in this report are described in the glossary in Appendix E. In particular note that the prototype lighting, discussed in this report, is based on the ceramic metal halide lamp (CMH). This is a development of the metal halide (MH) lamp in which an alumina lamp capsule replaces one made from quartz, allowing more aggressive salts to be used in the discharge at higher temperatures. TRL Limited 4 PPR043

15 2 The application of mesopic vision models to street lighting 2.1 Background The relation between street lighting level and accident rate is ill-defined, as is the interaction between vehicle lighting and street lighting. Nonetheless, a number of studies have shown that the provision of lighting on motorways can reduce accident rates. There is some evidence to suggest that light sources that produce a greater proportion of their output in the shorter wavelength (blue) region of the visible spectrum might be effective and acceptable at lower lighting levels. These sources have a broader spectrum than the high-pressure sodium lamps in common use at present and provide a whiter appearance and a more natural rendering of colours. It is known that the peak response of the human visual system shifts towards the blue region at and below the light levels currently used for street lighting. Utilising this effect may give greater visual effectiveness but with lower energy-consuming lighting systems. However, because of the complexity of the human visual system and the road-user s task, evidence is needed to confirm that reduced illumination will not have any adverse effect on road safety. Some research has shown that the shift in the visual system response affects the peripheral vision rather than the central visual field. The whole of the visual field is important to the road user, but the relative importance of the central and peripheral fields is unclear. The effect of the light spectrum on the detection of hazards has not been quantified. Therefore the efficacy of lighting tuned to the mesopic response of the human eye must be quantified for roadway and urban outdoor lighting schemes. The practical research at TRL described here was designed to achieve this by comparing aspects of visual performance relevant to the road user s task, under conventional and prototype lighting, in a controlled environment. 2.2 Objectives The first tasks to be undertaken included a review of human visibility models under mesopic conditions, the theoretical comparison of ceramic metal halide (CMH) and high-pressure sodium (HPS) lamps and the criteria for optimal luminance distribution on road surfaces and surrounds including environmental aspects. Practical research into the visual performance of road users under street lighting was also reviewed Human visual response to mesopic light levels A particular goal of the project was to reduce the luminous flux requirement of lighting by utilising the observed shift to shorter wavelengths of the peak sensitivity of the human visual system at low light levels. This must be achieved while still providing illumination satisfactory for the needs of all road users. HPS lighting provided the reference to which the CMH lighting developed in this project was compared. The CMH lamp produces a broader spectrum than the HPS lamp, with more of the energy at shorter wavelengths. This is clearly demonstrated by the comparison of Figure 2.1 and Figure 2.2, which show relative energy spectra for one of the prototype CMH lamps and one of the standard HPS lamps respectively, as used in the track experiment at TRL. It is therefore possible that, for perceived equal illumination, the CMH system need produce less luminous flux when measured conventionally, using the photopic human sensitivity weighting appropriate to daylight levels. However the human visual system is extremely complex and not fully understood. The parameters relevant to the visual tasks of road users and their safety must be assessed and appropriate experiments performed to demonstrate the relative performance of the reference and prototype lamp systems. TRL Limited 5 PPR043

16 Irradiance [W/(m².nm)] Wavelength (nm) Figure 2.1 Energy spectrum of a 150W CMH CCT=4000 K lamp Irradiance [W/(m².nm)] Wavelength (nm) Figure 2.2. Energy spectrum of a standard 150W HPS lamp TRL Limited 6 PPR043

17 2.3 The human visual system The human visual system comprises the eye, optic nerves and the visual cortex of the brain. The perception of brightness, contrast, shapes, colour and movement is a synthesis of the response of all these components. Studies of the human visual system have been hampered by their interdependence making it difficult to model the overall response from the measured responses of each part of the visual system. The factors important to vision under street lighting are discussed below The retina The retina is at the inner surface of the eye and contains the receptors generally known as rods and cones, a description arising from their shape. The rods are considered responsible for vision in the scotopic and mesopic zones (see Section and Appendix E for zone definitions) and monochromatic. The cones are considered responsible for vision in the mesopic and photopic zones and for colour vision. The geometry of the lens and retina is such that, in the horizontal plane, the visual field for both eyes together subtends about 180º. The peripheral retina receives light from the largest part of this field. Different authors have used varying terms for the areas of the central retina, which can be confusing. The central part of the retina, the fovea centralis, which is about 1.5mm in diameter and covers a visual field of 7º to 10º. The fovea contains the foveola, which is about 0.35mm in diameter and covers about 1.5º at the centre of the visual field. The foveola is not centred on the optical axis of the lens but is 5º horizontally and 2º vertically from it. The macula is 5.5mm in diameter and overlays the fovea and foveola, covering a visual field about 25º. This appears to be yellow, and may be an optical filter, but its function is not fully understood. The blind spot or optic disc lies 16º from the foveola and covers a very large area of 5º horizontally and 7º vertically. The spatial distribution of the photoreceptors over the retina is very non-uniform as is shown in Figure 2.3. The cone density is very high in the foveola, which accounts for the high visual acuity of the central 2º of vision. There are no rods in this zone. Conversely, there are relatively few cones in the peripheral field, which is mainly populated by rods, and this is considered to account for the relatively poor visual acuity (sharpness of an object) in this region. Visual acuity is reduced by chromatic aberration of the lens of the eye. This results in the focusing distance of short wavelength (blue) light being shorter than that of long wavelength (red) light, preventing a single, sharp focus. This effect becomes more significant at the large pupil sizes associated with low light levels. For this reason, Van Bommel and De Boer (1980) concluded that for equal visual acuity the luminance of a road surface under high-pressure mercury lamps would need to be 50% higher than under low-pressure sodium and under HPS would need to be 25% higher. If acuity was the only criterion of importance to the road user, then this could favour monochromatic light sources over broad-spectrum sources. However, Murray et al (1998) report that although acuity reduces with light level, it is unlikely to be related to everyday demands on the eye such as when driving. Visual reaction time is considered to have a more direct influence Adaptation to different light levels The adaptation process is complex and not fully understood. The change in pupil size, which is controlled by the rods, has only a minor role in adaptation. According to Schreuder (1998), the ratio of maximum to minimum pupil area is about six, whereas the total adaptation range is about thirteen orders of magnitude. The remainder of this range is achieved by complex electrical and chemical changes in the retina and the neuronal system. The pupil is at close to its minimum size throughout the photopic range. Through the mesopic range the pupil dilates so that at scotopic levels (see Appendix E glossary) it is fully dilated. The opening of the iris has a detrimental effect on visual acuity by increasing spherical aberration and also by reducing the effective depth of focus of the eye. These are the main causes of the phenomenon known as night myopia. TRL Limited 7 PPR043

18 Published Project Report Version: 1.0 Figure 2.3. Distribution of rods and cones in the retina (Reproduced from: It is well known that adaptation from light to dark is a much slower process than from dark to light. Full dark adaptation from photopic to scotopic levels can take at least 30 minutes whereas, in the opposite direction, the process only takes from a few seconds to a minute, depending on the change required. If the lighting level changes faster than the adaptation process can accommodate it then an adaptation defect exists and the observer will either be dazzled by light which is too bright, or if the level is reducing, unable to detect objects below the current detection threshold. It is therefore important to avoid subjecting road users to rapid changes of illumination Colour vision In the conventional model of vision, colour perception is only possible at mesopic and photopic levels and is a function of the cones. There are three types of cones sensitive to different wavelengths of light. Due to extreme difficulties in determining these wavelengths, researchers have arrived at various different values for the wavelengths of peak sensitivity. Published values vary at least from 440nm to 498nm for the S-cones, 535 to 545nm for the M-cones and 564nm to 580nm for the Lcones. In the conventional model of vision, luminance information is derived from the summation of the signals from all three of these channels, while chrominance (see Appendix E, Glossary) information is derived from the differences between them. The brain, using these chrominance signals then computes the sensation of colour. There are various factors that make measurements difficult. Firstly, the visual system adapts to colour. This can be demonstrated by staring at a coloured field for a short time and then looking at a white surface. A complementary colour is seen. This makes colour and brightness comparisons difficult and is the reason for the development of flicker-photometry (see Appendix E, Glossary). Another factor is colour constancy, which is the ability of the vision system to recognise objects as the same colour, even under very different illumination spectra. Fulton (2002) has published, on the Internet, a draft of voluminous and comprehensive research work on human vision. This presents an alternative, and perhaps controversial view, based on a re-analysis of published work. Fulton proposes that much published work gives evidence of four types of receptor in the human eye, which have peak sensitivities at 342nm, 437nm, 532nm and 625nm. They are designated UV (ultraviolet), S (short), M (medium) and L (long). The UV sensitivity in humans is suppressed by absorption in the lens and aqueous humour (see Appendix E, Glossary). The visual process involves multiplication of signals (logarithmic summation) not summation, which generate two sensation peaks where the receptor responses overlap (494 and 580nm). The rods are not achromatic (see Glossary) receptors, but a different form of cones. At scotopic light levels the L (red) receptors no longer TRL Limited 8 PPR043

19 function. This explains the shift to a lower wavelength composite peak in the scotopic zone. There are no receptors in the eye (the rods of the conventional model) with a peak sensitivity at the wavelength of 507nm (the peak sensitivity of the rods in the conventional model), which is the composite sensitivity of the M and S receptors. The receptors are sensitive to photon density rather than energy (less photons are needed for a given energy at short wavelengths). Colour sensitivity decreases through the mesopic zone and is absent in the scotopic zone. In particular the monochromatic rods and colour cones model is questioned. Fulton shows how the Commission International d Eclairage (CIE) curves are greatly smoothed versions of the true response, based on inadequate measurement capabilities in the 1930's. The main significance of Fulton s work for the current study is that the visual system s sensitivity to long wavelength light reduces in the mesopic zone. Lighting that stimulates the short and medium wavelength receptors is therefore likely to be more effective in this zone than the longer wavelengths Summary The factors important to vision under street lighting are the responses of the retina, including adaptation to different light levels and colour perception. Adaptation is a complex process and not fully understood. It is the process by which the human visual system adjusts to accommodate a wide range of brightness. The opening of the iris has a detrimental effect on visual acuity by increasing spherical aberration and also by reducing the effective depth of focus of the eye, preventing a single, sharp focus. However, visual reaction time is considered to have a more direct influence on driving than visual acuity. The key receptors of the human eye have peak sensitivities at 437nm, 532nm and 625nm. They are designated S (short), M (medium) and L (long). Colour sensitivity decreases through the mesopic zone and is absent in the scotopic zone. Lighting that stimulates the short and medium wavelength receptors is therefore likely to be more effective in this zone than the longer wavelengths. 2.4 Light and measurements Information about the zones of vision, the spectral sensitivity of the visual system and the change in visual response curves when departing from the photopic zone are discussed in this section Zones of vision Table 2.1 illustrates the approximate boundaries between the zones of vision in terms of luminance sensitivity. The photopic zone roughly equates to daylight levels. In this zone the visual system is conventionally considered to depend on the composite response of the cone sensors of the retina. The scotopic zone roughly equates to starlight levels. In this zone, the visual system is considered to depend only on the response of the rods. The mesopic zone covers the transition between the photopic and scotopic zones in which both rods and cones are active. Adaptation is the process by which the human visual system adjusts to the wide range of brightness that it can accommodate, ranging from 10-6 cd/m 2 to 10 7 cd/m 2. Road surface luminance levels under street lighting generally lie in the region between 0.1cd/m 2 and 2.0cd/m 2. This range lies at the upper end of the mesopic region, where cone sensitivity is still high and the contribution from the rods is expected to be small. TRL Limited 9 PPR043

20 Table 2.1. Ranges of human vision zones Upper (cd/m 2 ) Lower (cd/m 2 ) Wavelength of max sensitivity (nm) Equivalent colour Photopic (cones) > Green/Yellow Mesopic (transitional) Transitional Scotopic (rods) Blue/Green Colour vision is considered to be a function of the cones, with the rods being sensitive to luminance only. This implies a loss of colour sensitivity as the luminance decreases through the mesopic zone, becoming achromatic in the scotopic zone. The wavelength of peak sensitivity in the scotopic zone is then due to the rods alone. This is discussed further in Section The spectral sensitivity of the visual system The human visual system is sensitive to wavelengths of the electromagnetic spectrum between about 380nm and 780nm (Figure 2.4). However the sensitivity is not uniform to all of these wavelengths. Photometry is the science of light measurement in a manner proportional to human visual response. Figure 2.4. An approximate representation of the colours of visible spectrum Schanda et al (2002) provide a recent review of photopic luminous efficiency functions. The standard weighting curve used to represent visual sensitivity was defined by the Commission International de l Éclairage (CIE, 1924). The curve represents the spectral response of a standard observer based on experimental observations, using a range of subjects. This curve was defined only for photopic levels, for young people, and more critically only for the 2º central visual field. It was derived from brightness-matching experiments using monochromatic point sources of light and flicker-photometry (See Appendix E - Glossary). All photometric instruments are calibrated according to the CIE weighting curve (see curve on Figure 2.5, CIE Photopic 1924 ), known as the V(S) curve, which has a peak at 555nm. It was recognised as long ago as 1951 that this curve is deficient in that it under-represents the sensitivity of the visual system to the short wavelength end of the visible spectrum (Schanda et al, 2002). The CIE V M (S) curve (see curve on Figure 2.5, CIE Vm Voss-Judd ) represents an attempt to improve the function at the shortest wavelengths. The scotopic luminous efficiency function was defined by the CIE (see curve on Figure 2.5, CIE Scotopic 1951 ) and was also determined by brightness matching. It exhibits a major shift in the wavelength of peak sensitivity from the photopic peak of 555nm to 507nm. TRL Limited 10 PPR043

21 1 Normalised relative sensitivity Wavelength (nm) CIE Scotopic 1951 CIE Photopic 1924 CIE Vm (Voss-Judd) Figure 2.5. CIE luminous efficiency functions Despite its shortcomings V(S) is considered a useful measure of foveal luminance. The latter is also well-correlated with measurements of sensitivity, for explaining acuity and reaction time. However, it has long been known that the photopic spectral sensitivity for fields larger than 2º differs from V(S). The CIE has considered a V 10 (S) function for 10 off-axis sensitivity but no standard has resulted. Abramov and Gordon (1977) found that, at a constant photopic level, the sensitivity peak 45º from the fovea shifts to shorter wavelengths (close to the peak of the scotopic curve). Schanda et al (2002) state that there is much to be resolved in the area of large visual fields under photopic conditions Measurements under conditions departing from the photopic zone Because nearly all photometric instruments are standardised to the photopic CIE V(S) curve, the measurements that are made should be reasonably comparable between instruments, depending on the accuracy of their compliance. However, photometric measurements are not a reliable guide to the performance of the visual system particularly at the levels used in street lighting, for a number of reasons. More importantly, the shape of the visual response curve changes throughout the mesopic zone, in a complex and poorly defined manner. It is also known that the perceived brightness of polychromatic light cannot be derived from the simple addition of the monochromatic brightnesses used to derive the curve (Schanda et al, 2002). Photometric measurements are also shown to be inadequate when considering responses derived from more than the central (foveal) visual field. Schanda et al observed that strictly speaking, different photopic applications need different luminous efficiency functions. For a given visual parameter, such as acuity, even when only considering photopic levels, the appropriate function would depend on the spectrum of the light source and the size of the visual field. To accommodate the use of photometric measurements, Lewin (2000a, 2000b) proposed that supplementary correction factors (termed lumen effectiveness multipliers) may be defined for different viewing conditions. Unfortunately, it would appear that to cover all the variables involved in the road user s task (see Section 2.6.1), a large number of different factors would be required. The CIE (2000) recently reviewed ten different models of Mesopic vision by various authors. Ten photometric systems were tested. These had been proposed to the CIE as methods of assessing the relative brightness of lights across the entire range of human visual sensitivity. Four of the models considered 2º visual fields and six covered 10º fields. The tests used heterochromatic brightness TRL Limited 11 PPR043

22 matching. The report notes that photopic luminance does not correlate well with perceived brightness and that all the systems produced assessments of brightness superior to current CIE systems. However, the CIE has not yet accepted any system as a standard for the mesopic region. Unfortunately, brightness matching measures only one element of visual performance, which is not necessarily the most important to road users under street lighting. There is evidence that parameters such as visual acuity and target detection speed are more appropriate, and that the performance of the peripheral field is important to the latter. It is therefore likely that models based on measurements of these will be much more relevant to the tasks specified in the project. This topic is considered further in Section Summary The photopic zone roughly equates to daylight levels whereas the scotopic zone roughly equates to starlight levels. The mesopic zone covers the transition between the two zones. Road surface luminance levels under street lighting generally lie in the region between 0.1cd/m 2 and 2.0cd/m 2 at the upper end of the mesopic zone. The CIE weighting curve, V(S), is considered to be a useful measure of foveal luminance at photopic levels, and is well-correlated with measurements of sensitivity, for explaining acuity and reaction time. Photometric measurements from most instruments are therefore standardised to the V(S) curve. Unfortunately, photopic measurements are not a reliable guide to the performance of the visual system at the levels used in street lighting. 2.5 Literature review More than 70 documents were consulted in the process of the literature review. These were identified through searches of the OECD s International Transport Research Documentation (ITRD) database and the Internet, together with some material provided by other consortium members. The ITRD is managed by the TRL Information Service and is a bibliographic database containing citations to the worldwide literature from 1972, current research projects, and computer programs on all aspects of road research, transport, and traffic planning. Section describes the mesopic vision models found in the literature and the results of experiments to determine the effects of altered lamp spectra on vision are described in Section Mesopic vision models Hurden et al (2002) Hurden et al (2002) carried out a series of experiments for the UK Department of Trade and Industry and derived an empirical model of mesopic vision. The objective was to find a model with a wide range of industrial uses, which is not task-specific, and is based on detection and visual search problems. This represents a departure from the models based on brightness-matching examined by the CIE (2000) (see also Section 2.4.3). Brightness matching has not been successful in defining a mesopic vision model. Visual search was considered a better test of visual performance for most tasks. Because of this finding, the Hurden et al model could be more relevant to the task of the road user at night than brightness matching. Because detection in the peripheral visual field was recognised as important, it was also required that the model was not restricted to a single location in the visual field. In Phase 1 (Hurden et al, 1997), the experimental visual task used a 20 inch (51cm) CRT computer display that was used to display a small target among a field of similar distractors. The spectra of the display s phosphors and their luminance/voltage relationships were measured to allow the calculation of specific scotopic/photopic luminance ratios. The CRT display was viewed at a distance of 70cm TRL Limited 12 PPR043

23 giving a field of view subtending 29º horizontally and 23º vertically. This configuration allowed the use of both the peripheral and central fields in searching for the target. The measured variable was the mean search time for fixed targets in a fixed search field. The target was a ring with a missing segment (Landolt ring - see Appendix E - Glossary), while the distractors were complete rings. The computer-generated targets were displayed at defined photopic and scotopic contrasts with the background. These photopic and scotopic contrasts were achieved by the use of colour (an achromatic target has a scotopic/photopic contrast ratio of 1.0). The variables in the experiment were visual search time, adaptation level, target location, target photopic contrast, target scotopic contrast and chromatic difference, all to CIE standards. There were 15 subjects, male and female, aged between 20 and 50 years. These had no ocular abnormalities and normal visual acuity, using their spectacles where necessary. For each light level under test, 49 different combinations of scotopic and photopic contrast levels were presented. Search times were found to range from 0.5 to 4.0 seconds. The search time results were plotted as 3D surfaces. The x and y-axes were the photopic and scotopic contrast respectively and the search time was plotted on the z-axis. Analyses of the data set revealed that the visual search time depended on scotopic contrast, photopic contrast and background luminance. Asymmetry suggested that high negative mesopic contrasts gave shorter detection times. Other experimenters were also reported to have found this. It also appeared that photopic contrast was still a significant variable at 0.22cd/m 2 and even at the lowest adaptation luminance of 0.044cd/m 2. A new parameter "effective contrast" was defined, taking account of both target luminance and chromatic contrast. The effective contrast of a given colour or luminance difference is made equal to the contrast of an achromatic contrast which evokes the same response. In terms of search time the sign of the photopic contrast was not found to be significant but that of the scotopic contrast appeared to be significant. Contrast and adaptation levels both had significant effects on search times. An empirical model was fitted to the data, which ranged from the low photopic through mesopic to the high scotopic zones. The empirical model was able to account for 58% of the variance in the data. Differences in the subjects accounted for 21% of the variance, while only 2.4% could be attributed to colour differences. It was deduced that if the background and the target have similar radiance distribution, then the choice of the V(S) function is not critical for the measurement of contrast. Also, the photopic and scotopic (luminance) contrasts are similar in this case. It was thought that chromatic contrast might contribute to performance in the low photopic zone. The search time model resulting from this phase of the work had the form f ( C p Cs Lb ) ST =,, k e ( 2-1 ) where: ST = search time in seconds. k 1 is a baseline search time modified by a term whose exponent is a function of C p, C s and L b, C p = photopic contrast C s = scotopic contrast L b = background luminance 1 + In Phase 1, colour was used to achieve different scotopic and photopic contrasts. Phase 2 of the work is reported in Hurden et al (2002). In this phase chromatic differences were explicitly examined. The first experiment investigated the effect of colour on conspicuity. A computer display was again used to display a task in which the subjects were asked to compare the apparent brightness of a sequence of paired chromatic and achromatic targets. The results showed that, for any given chromatic stimulus, there is an achromatic stimulus of equivalent conspicuity. Under mesopic conditions, it can therefore TRL Limited 13 PPR043

24 be considered that conspicuity is a function of the same photopic and scotopic contrasts, chromatic difference and background luminance. Two subsequent experiments showed that performance, measured as search time, was predicted well by the generic achromatic performance model. Taskspecific, achromatic data can be used as the basis for determining performance at mesopic light levels. The following empirical 10-term polynomial was fitted to the data to give the achromatic match contrast. This model, a simplified version of a 19-term model, explained 84% of the variance of the data set whereas the full 19-term model explained 89% of the variance of the data set. C m k1 + k2. C p + k3. Cs + k4. Cs + k5. CD + k6.log10 Lb = exp ( 2-2 ) + k7. C p.log10 Lb + k8. Cs+.log10 Lb + k9. Cs.log10 Lb + k10. CD.log10 Lb where: C m = achromatic match contrast C p = photopic contrast C s+ = positive scotopic contrast C s- = negative scotopic contrast CD = colour difference in chromaticity space L b = background luminance in milli cd/m 2 k i = coefficients of model fit (tabulated in Hurden et al (2000)). The following empirical calibration curve, of a similar form to Equation 2-1 was found to be a good fit to the data over a range of background luminance levels: kcm ( t T ) e ST = T + 0 ( 2-3 ) where C m is the equivalent achromatic contrast and T, t 0 and k are fitted model coefficients, which vary depending on the background luminance. These are again tabulated in Hurden et al (2002). A number of calibration curves were required to cover the mesopic range. These calibration curves are specific to the target used in this task and for similar visual search tasks would be expected to vary with change in the parameters such as target size and presentation time. Different visual tasks would require determination of calibration curves specific to the task in question He et al (1997) He et al (1997) have published an experimental study of light source efficiency using reaction times. In the introduction to their report, evidence is presented that models based on reaction times are likely to be most appropriate in defining mesopic photometry for human tasks. They note that many studies have unsuccessfully attempted to define mesopic performance, mainly using brightness matching techniques. Few studies have investigated reaction time. Some studies have shown that the visual system uses different channels for different types of data. In particular, Kaplan and Shapely (1986) describe a magnocellular (MC) channel, which has fast response and is achromatic (i.e. a luminance channel), a parvocellular (PC) channel, which has better spatial resolution, and a chrominance channel (CC), which provides colour information but is slower. Experimental methods that invoke the MC channel are stated as likely to result in luminance functions that follow Abney s law of additivity, which is essential to photometry. Methods tapping the PC channel result in severe departures from linearity. TRL Limited 14 PPR043

25 Flicker-photometry was used to define V(S), which therefore probably isolates the achromatic MC channel, being the only channel with a fast response. Hough (1968) and Hough and Ruddock (1969) pointed out that, at mesopic levels, this technique may measure the response of only the cones or rods, depending on the light level and the position and size of the image. However, the combined response is believed to be important to visual perception at mesopic levels. Heterochromatic brightness matching has been used in most mesopic vision studies. This causes large variations, probably because both channels are used and the additivity assumption is not valid. Reaction time studies probably only tap the MC channel, and lead to dependence only on luminance. Thus reaction time techniques show promise for defining mesopic photometry. The experiment described used a light box in which a shutter was used to project a 2º target spot with its contrast fixed at +0.7 relative to the background. This target size was small enough to be imaged by the fovea when seen axially and large enough to be easily detected when seen off-axis. Reaction times were measured to the appearance of the target both on axis and 15º off-axis. A pencildrawn circle subtending 3º in diameter at the viewing position defined the axis. The light was either HPS, using a 400W lamp, or MH, using a 175W lamp, and the background level was varied from to 10 cd/m 2 in eight steps. Exposure was for up to two seconds if there was no response. Detection time was measured electronically between the opening of the shutter and the operation of a switch by the observer. Three observers were used who had normal vision. Two were males aged 28 and one was female aged 31 years. It was found that the type of light-source did not affect the reaction time (RT) when the target was onaxis. On-axis reaction times were found to be explained by an equation of the form: RT = a / L b + c ( 2-4) where: L = background luminance a, b and c are coefficients, which may vary with target position, light spectrum and observer. Off-axis there was no discernible difference in reaction time above 1 cd/m 2. Below 0.3 cd/m 2 reaction times for HPS became progressively longer. This result agrees with other studies. Brooke (1951) showed a divergence at 0.5 cd/m 2 and Hecht and Shlaer (1936) showed branching below 0.8 cd/m 2 for seven monochromatic wavelengths. Guided by this 0.6cd/m 2 was chosen as the level, below which the off-axis data diverged, when fitting curves to the data. This implies that, for detection of the 2º target 15º off-axis, there is no contribution from rods above 0.6cd/m 2. Below this, the rods contribute progressively more as the level drops. Table 2.2 Table 2.2. Off-axis luminance ratio of MH to HPS lamps (He et al, 1997) Luminance level (HPS) (cd/m 2 ) Average MH to HPS luminance ratio (R(L)) TRL Limited 15 PPR043

26 shows the ratio of average off-axis MH to HPS reaction times [ R(L) ] for the range of background luminance covered. A model of mesopic vision is proposed, which is a simple linear combination of rod and cone responses. This utilises the CIE V 10 (S) (representing the large-field cone response) and V'(S) (representing the scotopic rod response) using a proportional factor x(l), which is derived from R(L). This model assumes the validity of Abney s additivity law in the mesopic region. V m ( ) = k ( L) [ x( L) V ( ) + ( 1 x( L) ) V ( ) ] 1 10 ( 2-5 ) Here k 1 is a normalisation constant to ensure that the maximum value of V m (S) is unity. For the lamps used in the experiment the following relationship was derived for calculating x(l). x ( L) = 1 R L R L ( 2-6 ) [ ( ) ] [ ( ) ] For any R(L) in Table 2.2 the corresponding x(l) can be obtained from equation 2-5 and the mesopic luminous efficiency function computed using equation Fulton (2002) Fulton (2002) has proposed an interesting and as yet controversial new model of vision, (referred to in Section 2.3.3) based on an extensive study of research evidence, in which the luminance information is derived from the logarithmic sum of the signals generated by the blue (S), green (M) and red (L) photoreceptors. The chrominance information is carried on two channels designated P and Q (see Figure 2.6). The P channel signal is the logarithmic difference between the S and M signals and the Q channel is the logarithmic difference between the L channel and the M -channel. The generic equation for the luminance signal (R- channel) is a = log k a + log k a + log k a (2-7) where: a t () k x t ( ) ( ( )) ( ( )) ( ( )) is the luminance signal, are a set of constants relevant to the adaptation zone s s a x () are the individual spectral responses of the photoreceptors By inserting the appropriate set of coefficients, k, the full range of adaptation can be covered. Appropriate coefficients for the photopic response are k s =50, k m =1000 and k l =30. Through the mesopic zone the sensitivity of the L-channel (red) falls. This is represented by the value, k l, which reaches a value of less than one in the scotopic zone. Knowledge of the relationship between the mesopic adaptation level and the coefficient of the L-channel would therefore provide a theoretical model of the response in the mesopic zone. In Fulton s model of vision, the regions of adaptation (see Table 2.1) are characterised as follows (based on Fulton, 2002): TRL Limited 16 PPR043 m Photopic region The operating region of the visual system where the adaptation amplifiers of the individual spectral channels are all operating at amplification factors greater than 1.0, but none have reached their maximum gain. Mesopic region The operating region below the photopic region characterised by the L- channel adaptation amplifier operating at full gain. The L-channel signal decreases more rapidly than in the M- and S-channel signals with decreasing illumination levels. It is a region of decreasing saturation in the perceived colours of objects due to a decreasing signal level in the P & and Q chrominance channels relative to the threshold level The transition between the scotopic and photopic regions is smooth, albeit not linear. The L-channel performance rises according to a square law relationship. m l l

27 Photoreceptors Log difference P- channel Blue (S) Chrominance channels Green (M) Log difference Q- channel Red (L) Weighted log sum R- channel Luminance channel Figure 2.6. Simplified signalling block diagram of the human visual system proposed by Fulton (2002) The adaptation amplifiers in all three spectral channels are operating at maximum gain at the transition between the scotopic and mesopic regions. Their gain does not begin to reduce until the transition region between the mesopic and photopic regions is reached. In the mesopic region, both the gain coefficients and the RMS noise threshold associated with each spectral channel are changing with stimulus level. Under these conditions, the gain coefficients maintain a fixed relationship with each other, however, the quantum noise is different in each spectral channel and the quantum noise in the L-channel exhibits a more complex relationship than it does in the other two channels. The relationship between the gain coefficients is typically k s : k m : k l = 50:1000:1 at the interface with the scotopic region and k s : k m : k l = 50:1000:30 at the interface with the photopic region, when including the absorption of the lens group. Scotopic region The lowest operating region of vision characterised by all adaptation amplifiers operating at full gain but a complete absence of L-channel sensitivity in the spectral response of the eye. A region of achromatic vision due to the signal level in the P and Q chrominance channels falling below the threshold level. In the scotopic region, the adaptation amplifiers in all three spectral channels are operating at maximum gain, but the incident radiation level is so low that the gain coefficient of the L- channel is negligible relative to the coefficient of the M-channel. This is due to the square law term in the L-term in the luminance summation equation. Under these conditions, the S- and M-channel gain coefficients maintain a fixed relationship with each other and this relationship is typically k s :k m :k l = 50:1000:(<1) when including the absorption of the lens group Experiments to determine the effects of altered lamp spectra on vision Akashi and Rea (2001) Akashi and Rea (2001) describe an experiment in which peripheral visual detection under both HPS and MH light at mesopic levels was measured, with and without opposing headlight glare. The experiment aimed to recreate the driving task more closely than in others reviewed here. It is stated that, at night, peripheral vision is used to detect hazards such as pedestrians or animals that are about to cross the roadway. The effectiveness of both central (foveal) and peripheral vision is affected by glare from streetlamps and opposing headlamps. There are currently few studies of the effect of glare effects on peripheral vision. The paper refers to published work by He et al(1997), Lewis (1998 and 1999) and Bullough and Rea (2000) for evidence that off-axis detection is better TRL Limited 17 PPR043

28 under MH lamps than HPS. However, it is also noted that other researchers have indicated that short wavelength light might be perceived as more glaring. The experimental work used an otherwise unlit parking lot. Five 3.66m (12 ft) high poles were erected at 20m intervals along a 3m wide, 80m long lane demarcated by cones as shown in Figure 2.7. Figure 2.7. Layout of experiment (reproduced from Akashi and Rea, 2001) Each pole supported a cylinder containing two luminaires. These contained a 150W HPS lamp and a 175W MH lamp and the cylinder was rotated to bring each luminaire into approximately the same position. Each luminaire was fitted with a mesh screen to produce 5.5 lux average illumination at the pavement surface. The absolute uniformity (e min /e ave ) was 0.15 for the HPS lamp and 0.11 for the MH lamp. Subjects sat in the left seat of a car located in the right-hand lane. The car was fitted with halogen headlamps. An opposing car was positioned 8.9 m from the subject car, in the opposite lane. The opposing car was also fitted with halogen headlamps, which produced an illuminance in the vertical plane of 2.4 lux at the driver s left eye. The detection targets were mounted 10m away and were 15º and 23º off-axis from the driver s viewpoint. These targets comprised 25cm x17cm LCD shutters mounted on the front face of boxes painted black inside and subtended about 1.35º x 0.91º at the driver s viewpoint. The shutters were switched from 0% to 20% reflectance in 0.25 sec by a timing system. The text states that the photopic luminances of the targets were maintained at 0.33 cd/m 2. However, Table 3 of the paper states 0.06 cd/m 2 with the headlamps off. Their contrast with the background was fixed at 0.54 for both HPS and MH light sources. For each observation the driver fixated on a distant LED target and, as soon as a change in a peripheral target was detected, touched a pressure-sensitive switch connected to the timing system. Reaction times were measured with both lamp types and with opposing headlamps on and off. The headlamps of the driver s vehicle were on for all conditions. Some light from these headlamps fell on the targets in addition to the illumination from the street lamps. While this contributed to the realism of the experiment, this light would have interfered with the comparison of the lamps under test. The footnote to Table 3 of the paper states that this resulted in a total luminance of 0.27 cd/m 2 for the 15º off-axis target and 0.13 cd/m 2 for the 23º off-axis target (This does not agree with the 0.33cd/m 2 quoted in the text). Eight subjects participated, aged from 22 to 39 years, with claimed normal vision. Each subject undertook observations of eight experimental conditions with 10 repetitions of each. These were randomised with 20 additional presentations in which no target was presented. Half of the observations were made with the HPS lamp and half with the MH lamps. The other variables were the target position and whether the glare was on or off. Although the table of variables includes Forward headlight: off or on the results presented do not mention effect of forward headlamps. The discussion also suggests that the headlights were always on. There were large differences in the reaction times of the individual subjects, which led to large TRL Limited 18 PPR043

29 standard deviations in the observations of each condition. The results were normalised by the subject s shortest reaction time (which was always for the MH/15 º/glare off ) condition. This reduced the standard deviations found from the data. The means of subjects reaction times are given in Table 2.3 and the reaction time ratios relative to the MH/OFF condition in Table 2.4. Table 2.3. Reaction times in ms Condition Target HPS/GL HPS/OFF MH/GL MH/OFF 15º º Table 2.4. Relative reaction times Target HPS/GL HPS/OFF MH/GL MH/OFF 15º º The results showed slightly shorter, but statistically significant, reaction times under the MH lamp than under the HPS lamp and slightly shorter reaction times for the 15º targets. Opposing headlight glare did not affect the reaction times for the 15º targets but it did for those at 23º. However, this result interacts with the effect of the forward headlamps, which illuminated the 15º targets more than the targets at 23º. This study emphasises how the results obtained in a realistic situation can depart from rigorously controlled experiments where most parameters are held constant. Although a slight improvement under the MH lamp compared with the HPS lamp was detected, it was much smaller than other studies might predict. Larger numbers of moving vehicles and other extraneous light sources would be bound to have a greater masking effect in most real-life situations. It is encouraging that reaction times were not badly affected by the opposing headlamp glare, as this is a potential obstacle to the luminance reduction aim of the NumeLiTe project Van Derlofske et al (2001) Van Derlofske et al (2001) have experimentally determined that high intensity discharge (HID) headlamps provide improved visual performance compared with halogen headlamps. Because these headlamps have some similarities in spectral characteristics to MH lamps, the findings are of interest to the current study. However the HID lamps also exhibit higher off-axis intensities than halogen headlamps and the effects of the shorter wavelengths in the spectrum could not be separated from the higher intensity. The experiment was performed using a visual task in which detection times for peripheral targets were measured. Two halogen headlamp types and one HID type were compared. This full-scale experiment was conducted on an unlit disused bituminous-surfaced runway at night. The observer sat in a stationary vehicle, in front of which were mounted the lamps to be tested. A tracking task fixation target was set 15m away on the subject s sight line and stimulus targets were placed 60m away at five degree intervals from 7.5º (driver s side) to +17.5º (passenger s side) (Figure 2.8). TRL Limited 19 PPR043

30 Figure 2.8. Experimental geometry (reproduced from Van Derlofske et al, 2001) The stimulus targets were 17.8cm (7in) square grids of 12.7mm (0.5in) flip-dots, which could be simultaneously switched from white to black within 20 milliseconds. Neutral density filters placed in front of the targets were used to provide two levels of contrast without spectral shift, designated high and 50%. The illuminance was measured on each target for each headlamp type (Figure 2.9). Figure 2.9. Target illumination (reproduced from Van Derlofske et al, 2001) Twelve subjects participated. Six were aged under 30 years and six over 50 years. All had 20/20 vision, with their usual correction where needed. Six targets, four presentations per target, two contrast levels and three headlamp types resulted in 144 observations per subject, divided between six sessions. The test sequences were randomised to remove order effects. The subjects responded to the detection of a target change by releasing a button. Any reaction time greater than one second was counted as a miss. Reaction times consistently increased with increasing angle and decreasing illumination. Low contrast also increased this effect. The older subjects had longer reaction times for both types of lamp than the younger subjects and this effect increased as the peripheral angle increased. However, the relative magnitude of this change was similar for both age groups. The HID lamps gave shorter reaction times, particularly at large angles, than the halogen lamps. The results of the study indicate reaction time benefits for HID lamps, but the effects of illuminance and spectrum could not be separated. It is also noted that the higher intensity of HID lamps may give rise to greater glare for opposing traffic. TRL Limited 20 PPR043

31 Bullough and Rea (2000) Bullough and Rea (2000) provide a study that used computer simulation to present a driving task and measure visual performance. The study was designed to further investigate the findings of He et al (1997), which covered both central and peripheral detection at mesopic and low photopic light levels. In that study, which used a projected circular spot of light against a plain illuminated background, levels above 1cd/m 2 HPS lamps were 20% more efficacious than MH lamps on- and off-axis. At 0.1 cd/m 2 MH lamps were up to 60% more efficacious in terms of 15º off-axis peripheral target detection, but HPS lamps were still 20% more efficacious than MH for on-axis target detection. This new experiment used a computer display projected on a 2.4m x 1.8m (8ft by 6ft) screen. This display subtended about 40º horizontally and 30º vertically at the observer s eye. Four discrete simulated driving tasks were created using proprietary game software. The projector used an MH lamp and combinations of colour and neutral density filters were interposed to control the brightness and spectral content of the projected light. Four different spectral power distributions (SPD) with the scotopic/photopic ratios (the ratios of measurements made with scotopic and photopic weightings, respectively) shown in Table 2.5 were used. In successive trials the illumination level was adjusted so that the time-averaged luminances, measured with a photopic-weighted meter, on the projected roadway (at centre of display area) were 3, 1, 0.3 and 0.1 cd/m 2. Table 2.5. Luminance levels for each experimental condition (reproduced from Bullough and Rea, 2000) Spectral Power Distribution (SPD) Scotopic/ Photopic ratio Scotopic luminance for photopic level (cd/m 2 ) Subjective appearance HPS Orange Red Pink MH White Blue Blue It is noted that at mesopic luminances the pupil is at maximum size, so this factor would not interfere with the illumination of the retina. In a second experiment, a peripheral target device was placed in the lower right corner of the display. This subtended 2.4º by 3º and was located 18º from the centre of the screen. Darkening the area of the display using a luminance contrast of 0.78 created the target. It could be made to appear at random for intervals of 0.5 second. Whether or not this was detected was recorded. Two simulated racetracks were used in the driving tasks, which were also mirrored to produce two more. In the first experiment, eight subjects took part, aged between 18 and 31 years, with normal acuity and colour vision. A total of 64 conditions were used comprising four light levels, four SPDs and four tracks. In the second experiment, six subjects took part aged between 24 and 38 years. A total of 16 combinations were used, covering the four light levels and SPDs. Statistics calculated for each run were driving speed, crash frequency, brightness rating and, in the second experiment, peripheral object detection. Speed increased significantly with increasing background luminance but there was no significant effect of different SPDs at any luminance. Crash frequencies reduced with increasing luminance, but again there was no significant effect of different SPDs. A subjective brightness rating was also affected by the different SPDs. The peripheral detection experiment showed a reducing percentage of misses with increasing background luminance, but the detection improved as the scotopic/photopic (S/P) ratio increased. For a common level of about 37% misses the HPS and MH photopic luminances were 3 cd/m 2 and 0.1 cd/m 2 respectively. This gave an unexpectedly high relative effectiveness ratio of 30 for MH to HPS lamps, given an S/P ratio of only TRL Limited 21 PPR043

32 2.78. There is also a clear indication that the percentage of misses was significantly lower at the higher S/P ratios for a given SPD even at the 3.0cd/m 2 background level. This suggests that the rods are still contributing to peripheral target detection at levels conventionally considered to be photopic. The results would be explained if the driving task, with no peripheral hazards in the scene, was an onaxis task, and central vision was not affected by different SPDs. The peripheral detection task was definitely off-axis and showed a much stronger effect of different SPDs at mesopic levels than would be expected from previous work. It is suggested that this could be because the previous work involved static scenes (i.e. less visual noise ) and this work used a moving background, representing a greater noise level. It is therefore possible that the MH light is more effective in situations near threshold levels. This work reinforces two ideas: That the basic driving task (following the road) is a function mainly of central vision and that this is little affected by changes in the SPD of the lighting. That the peripheral field is very sensitive to the SPD of the lighting and that this is important for detection of hazards, especially those near the edge of the road Effects of reduced luminance on visibility Bacelar and Beaucamp (2001) Bacelar and Beaucamp (2001) describe an experiment to determine the effects on visibility of reducing street lighting illumination. A 7m wide test road at CETE, Rouen was used. This was equipped with five dimmable HPS luminaires mounted at a height of 8m and spaced 30m apart. All the luminaires were on the same side of the road. The longitudinal and transverse uniformities achieved were 0.75 and Observers were placed 83 metres longitudinally from the position of luminaire 2, as shown in Figure In this figure luminaires 2 and 3 are on the right and distances along the road from the position of luminaire 2 are on the left. Between luminaires 2 and 3 the road was divided into a 4m x 7m grid. The bottom row of the figure shows the distances of the cell boundaries from the line of the luminaires. The pavement was surfaced with an R1 type coating, with a Q 0 of A single target was presented randomly at each of 10 selected points, as shown in Figure The target was a 20 cm diameter ball of diffuse reflectance of 0.2. Dimming of the lamps was used to set illuminance levels of 25%, 50%, 75% and 100% of maximum. At the 100% level, the average illuminance was 30.6 lux and the luminance at the observer s viewing position was 2.89cd/m 2. The eight observers, aged from 21 to 59 years, viewed from a standing position. For each trial they faced away from the scene while the target was placed. Two sound signals then initiated a turn to face the scene, and away again, respectively. This permitted a viewing time of between 200 and 500msec. Visibility was rated on a subjective scale from 0 (not visible) to 4 (good visibility), as shown in Table 2.6. TRL Limited 22 PPR043

33 Figure Diagrammatic layout of the experiment (Reproduced from Bacelar and Beaucamp, 2001) Table 2.6. Rating scale Rating Assessed Visibility 0 Not visible 1 Weakly visible 2 Visible 3 Satisfactory visibility 4 Good visibility The ratings were plotted against the measured visibility level (VL) (defined in Section ) of the target as shown in Figure It was considered that a rating of 2 corresponded with the minimum visibility required. For this experimental setup this correlated with a VL of 4 and corresponded to an illuminance of 15.3 lux and a luminance of 1.44cd/m 2. Previous model work (Lecocq, 1999) resulted in a minimum VL of 7 for a 20% reflectance target and a 350 millisecond observation time. The above therefore suggests that a lowering of the minimum visibility level might be safe. However, the observation time was not very tightly controlled in this experiment; the observers did not cover the oldest road-users and the observations were not made from within a vehicle, so the results need to be treated with caution. It would be very interesting to replicate this experiment with the addition of a comparison of CMH and HPS lamps. TRL Limited 23 PPR043

34 Figure Visibility of target (reproduced from Bacelar and Beaucamp, 2001) Collins et al (2002) Collins et al (2002) describe a trial of dimming on the relatively lightly trafficked M65 motorway in Lancashire. The traffic flow is more than vehicles per day on the busiest section. The lighting has been replaced by a dimmable system, controlled by traffic flow, along the 7-mile, dual two-lane section between Junctions 10 and 14. The new lamps are 150W HPS in full cut-off luminaires, controlled by a high-frequency electronic ballast and electronic dimming. The system is remotely controllable through a power line modem. The luminaires are mounted 12m high (10m on slip roads) on the existing columns. The spacing is not stated, but the whole scheme involved about 700 luminaires. Design of the new scheme was to CEN pren , Class ME2. This required minima of average luminance (L AV ) of 1.5 cd/m 2, absolute uniformity (U 0 ) of 0.4 and longitudinal uniformity (U L ) of 0.7. Table 2.7 shows the measured levels achieved with the old and new systems in practice. Table 2.7. Comparison of measured lighting levels in old and new schemes OLD NEW Lantern 180W SOX LTI WRTL Vectra 150W SONP/T Flat glass Carriageway Hard shoulder Slip roads L AV U 0 U L 1 U L 2 L AV U 0 U L 1 L AV U 0 U L 1 U L The full-load power consumption of the old system was kw whereas that of the new system is kw, representing a 24% reduction. Three lighting levels shown in Table 2.8 are being used in the trial. The level to be applied is controlled automatically by the traffic flow measured on the busiest section of the road. When switching occurs, the level is changed gradually. TRL Limited 24 PPR043

35 Application of the current traffic flow profile at the three lighting levels should result in a total energy saving in excess of 50%. The corresponding energy and cost savings are shown in Table 2.9. This scheme has also been the site of research by the University of Manchester Institute of Science and Technology into the effects of dimming on glare and driver comfort, using its ocular stress monitoring technique, as described in Section Table 2.8. Flow steps and lighting levels Vehicles per Hour Lighting level > % % < % Table 2.9. Predicted energy and cost savings Annual energy cost (kwh) Annual cost at 4.2p/kWh Original system New system alone With flowcontrolled dimming 637, , ,000 26,769 20,344 12, Effects of glare on visibility Cooper and Diver (2000) There are two types of glare - disability glare and discomfort glare. In the driving context, most research work has concentrated on disability glare, which reduces the ability of drivers to perceive objects in the road ahead. Discomfort glare is relatively little understood and whilst some work has been done on factors contributing to it, it is not treated specifically in BS5489 (10 parts - BSI, ), which attempts to limit disability glare caused directly by a lighting installation. It assumes that if disability glare is under control, then discomfort glare should be acceptable. Other than a lighting installation itself, the main glare sources are headlamps of oncoming and following vehicles, and vehicles rear lamps. Disability glare Disability glare can be described as the glare that impairs the vision of objects without necessarily causing discomfort (van Bommel and de Boer, 1980) The mechanism of disability glare is the scattering of light within the eye. A glare source produces an equivalent veiling luminance on the retina. This reduces the contrast of the scene and the ability to perceive objects. Vos (1984) concluded that disability glare was due to scattering in the eye rather than due to any inhibition of neural processes. The scattering occurs in the cornea, the lens and the retina. The scattering becomes greater with age, particularly in the lens of the eye. The glare effects are about three times as severe in an 80-year-old observer compared with a 20-year-old observer. In addition, older people have TRL Limited 25 PPR043

36 reduced contrast sensitivity, i.e. they will be less able to discern small luminance differences, although it is not clear whether this and the increased amount of scattering in the eye are two independent phenomena. Adrian (1968) also noted that older people experienced more glare because the structure of the lens and its spectral transmission change with age. Blue light transmission falls and the amount of stray light increases through greater scattering, particularly at shorter wavelengths. Adrian also noted that the effect of different glare sources was additive. Disability glare can be mitigated by: reducing the intensity of the glare source moving the glare source further from the line of sight increasing the adaptation level (Keck, 1983). Discomfort Glare Discomfort glare has been described as the glare that causes discomfort without necessarily impairing the vision of objects (van Bommel and de Boer 1980). Many road users experience glare from badly placed or very bright lights that cause discomfort. The level of discomfort ranges from mild irritation to extreme stress depending upon the lighting and the sensitivity of the observer. Discomfort glare does not necessarily interfere with the road user s visual performance, at least in the short term. The mechanism giving rise to discomfort glare is not wholly understood, but the degree of discomfort depends on the intensity of the light source, its position in the field of view and on the background luminance (see for example Schmidt-Clausen and Bindels, 1974). Some authorities have speculated that the sensation may originate in the pupil, which is richly invested with pain receptors (Fry and King, 1975). Most of the research into the possible physiological basis has been connected with the change in pupil size that regulates the amount of light entering the eye. Previous research produced conflicting results. Fry found that discomfort could be related to pupil size. Hopkinson and Petherbridge (1950) found that the pupil became noticeably unstable in conditions producing discomfort glare and that this instability increased as glare increased. However, Howarth et al (1993) did not find this effect. There was possibility that pain detectors in the extra-ocular muscles are the basis for the response. Sivak et al (1990) found that discomfort glare was task dependent. In a laboratory experiment they asked subjects to identify the location of a gap in an open square, and also to report the degree of discomfort produced by a glare source. Whether discomfort increased or decreased as a result would depend on the strengths of the relationships with task difficulty and non-uniformity of the visual field. This is an important consideration in relation to the driving task. Flannagan et al (1989) used a number of essentially monochromatic sources to look at the effect of wavelength on the level of discomfort glare. They found that the degree of discomfort increased not only with decreasing wavelength but also, to a lesser extent, at longer wavelengths (above about 600 nm). Older drivers experienced more discomfort glare than younger ones, except at the shortest wavelengths investigated. It was suggested that this latter effect might be due to yellowing of the lens with age. Sivak et al (1997) found that for a given target, discomfort glare depended on the illuminance at the pupil and on the duration of the stimulus. The authors noted that whilst the effect of illuminance was fairly constant over all subjects, the effect of duration was very variable. Sivak and Olson (1983) stated that both disability and discomfort glare were dependent on intensity, distance, angular separation from the line of sight and the adaptation level. Although disability glare was attributed to scattering, and hence was age dependent, the uncertainty in the mechanism of discomfort glare did not allow the authors to conclude that the effects were necessarily likely to increase with age. Adrian (1968) found that the level of discomfort glare depends upon the size of the source and also suggested that discomfort glare can occur at low luminance light levels. It is thought that discomfort glare TRL Limited 26 PPR043

37 was likely to be due to a range of effects having different causes, linked only by their association with excessive contrasts in the visual field. Recent work for TRL, carried out at UMIST (Cooper and Diver, 2000) on discomfort glare, showed that there was a strong electromyogram (EMG) response from headlamps used as a glare source but only for relatively short viewing distances. This was consistent with measurements of corneal illuminance. It was concluded that, at a viewing distance of 20m and beyond, provided they are optimally adjusted, average headlamps do not constitute a major glare source. Results indicated that at 20 m dipped headlamps produced no change in background EMG response in a subject sitting in a car facing the headlamp beam. There was a small change in EMG response (around 40%) when the headlamp was on main beam. Although this effect was statistically significant, this represented 'just detectable' discomfort glare. At the 15 m viewing distance, the effect is much stronger for the main beam. By extrapolation it can be concluded that this effect (200% change) constitutes a perceived glare rating between 'unacceptable' and 'intolerable'. The dipped beam produced only a small (20%) increase in EMG response at this distance. This work was carried out with stationary vehicles and no street lighting. The UMIST experiments also showed that as the luminance level of a glare source increases (and stimulus size falls), glare is also increased. Furthermore glare sources are cumulative, that is, two glare sources of equal size and photometric output, generate twice as much discomfort glare as one. Implications for road users It seems that disability glare is related to the intensity of the glare source; discomfort glare is also related to intensity for a given size of source, but is primarily related to luminance. From this, it follows that for a given intensity of light the relative importance of discomfort glare becomes greater as the size of the source is reduced. It would be expected that the discomfort glare produced by high-pressure sodium (HPS) luminaires would be greater than that associated with the larger low-pressure sodium sources, other things being equal. Similarly some new headlamp designs, which have a smaller effective area than conventional lamps, would be expected to cause more discomfort glare than the conventional designs. If the lights produced the same intensity, disability glare would not be affected (see Schmidt-Clausen and Bindels, 1974). Adrian (1968) compared disability glare and discomfort glare as a function of mean surface luminance. As luminance increased, disability glare was noticed first at low surface luminance, but Adrian stated that discomfort glare, dependent on source area, may be more important overall. At street lighting levels, the prevention of discomfort glare required more care than prevention of disability glare. An experiment was carried out on disability glare in which the luminance of a disc was varied until it was just visible, with and without glare. It was demonstrated that older drivers had to set the target disc luminance higher than younger drivers even without glare and they had to set it disproportionately higher to compensate for glare. Older drivers also took longer to recover from exposure to glare. Keck (1983) calculated the effect of glare on visibility for vehicles meeting on a two-lane road. Increasing the adaptation level of the observer can mitigate the effects of glare. Increasing road surface luminance can increase adaptation. The adaptation is generally governed by contributions from fixed lighting installations and the headlamps of the driver s vehicle falling on the road up to 60 metres ahead, i.e. prior to the section whose luminance is addressed by the lighting standard. The output of vehicle headlamps has to meet a standard that defines maximum or minimum light levels at a number of key points in the angular distribution of emitted light. One of these, the glare point, seeks to limit the illuminance at the eyes of oncoming drivers. The type of headlamp can be an important determinant of glare. A recent trend in headlamp design is towards headlamps with smaller effective areas. The beams of these are usually collimated by a series of lenses. They may therefore give rise to more discomfort glare unless the light output is carefully controlled. Headlamp mis-aim and particularly the effects of road gradient and curvature may become more important. TRL Limited 27 PPR043

38 High intensity rear lamps of a preceding vehicle are another source of glare. The intensity of the light is less than from headlamps, but they are likely to be closer to the driver and nearer the driver's line of sight. There is little in the literature about this Murray (2002) This paper introduces the concept of dynamic discomfort glare from non-uniform road lighting. This type of glare has been evaluated by measuring ocular stress using the electrical signals from electrodes placed around the observer s eye. In the laboratory a flicker rate of 1Hz (equivalent to that produced at 70 mph with luminaires at 30 m spacing) was found to produce the most discomfort to observers. In a field experiment, on the motorway described in Section , the vertical illuminance on the faces of observers, sitting in the front passenger seat, and their ocular stress were measured. It was found that lowering the sun-visor had a big effect in reducing the illuminance at the driver s eye position. As the vehicle progressed along the road the range was 1.5 lux to 40 lux without the visor. With the visor lowered the range was 1 lux to 2 lux (a tall driver would, of-course, receive some benefit from the top of the windscreen, without the visor). The ocular stress response was greatest on an unlit section of road. This was clearly due to glare from opposing headlamps. It was reduced by overhead lighting and further reduced by lowering the visor. A comfort index scale of is suggested. The results indicate that lighting levels should be greatest when flow and opposing headlamp glare are greatest Road user visibility models The current UK and proposed European standards for lighting design are based on either the specification of luminance of the road surface seen from the driver s point of view or of the illuminance provided by the lighting. Which is appropriate depends on the class of road and the type of traffic. In the luminance case, it is expected that potential obstacles will be darker than the background ( negative contrast ) and that the background is the road surface. This assumption is not valid in many cases. In particular the contrast between an object and its background will vary continuously as their position changes relative to the lamps. There may be positions where the contrast is close to zero. The illuminance-based standards are used in urban mixed pedestrian and vehicle situations where the luminance of the background is less certain. None of the standards attempt to account for spectral effects and all are based on photopic lumen measurements. It has long been recognised that an important factor in determining visibility is luminance contrast. Colour contrast may also be important but has not been investigated to the same extent. The current standards and visibility models based on luminance contrast that have been proposed as the basis for alternatives are described below British and European Standards Current British and European lighting designs have the general aim of providing illumination to horizontal and near-horizontal surfaces while minimising the illumination of vertical surfaces. The luminance of the road surface and its method of measurement are specified, over an area of road surface some distance ahead of the driver. The intention is that objects, which must be detected 60 m or more away, are revealed in silhouette. It is recognised that the contrast of objects against their background will vary and that there will be some positions in which parts of an object will have the same luminance as the background. However, not all parts of an object will have the same reflectance, the background luminance will not be uniform and relative movement will ensure that the background changes. It is also recognised that objects on the footway or at the side of the carriageway should also be visible. TRL Limited 28 PPR043

39 In urban centres, with mixed vehicle and pedestrian traffic, and where the effective background to objects will often not be the road surface, minimum illuminance levels are specified instead of luminance of the road surface. More detail is given in Appendix B Small target visibility Southall and Shearlaw (1997) have reviewed lighting and visibility models for small targets. Successive models from 1938 to 1989 introduced additional variables. Adrian s model (1989) includes seven variables. However, none of these models include lamp spectral effects or traffic flow. Available observation time for the detection of targets is considered to be typically 0.2 seconds. Viewing time has to increase to detect smaller targets. Adrian (1989) estimated that an average visibility level (VL) of between 10 and 20 is needed for safety. Keck and Stark (1987) evaluated models in the field using a variety of geometric solids as targets. All the models over-estimated target visibility under positive contrast. The model presented in CIE 19/2 (1972) was the best, under negative contrast, while Gallagher and Maguire (1975) and Adrian over-estimated. There was little difference between flat and 3-D targets except directly under a luminaire. The Field Factor of Narisada (1995) is described; unexpected targets need greater luminance to be seen at the same distance as expected targets, which has implications for experimental design. However in this experiment, data were collected for headlight illumination only. The debate continues over the best approach - CIE high luminance, high uniformity, or Adrian s STV model. CIE Technical Committee 4 has recently produced a draft document (CIE, 2002) presenting a collection of material on visibility design for roadway lighting. Reduction in the level of night-time accidents is the main purpose of street lighting. However, it is reported that research has shown that design based on the visibility of small targets can be more effective than the traditional luminance and illuminance concepts. Studies in the USA are quoted as having demonstrated that the levels of lighting under current standards do not correlate well with day/night accident ratios. The small target visibility (STV) concept is based on the average visibility level (VL) of a large number of small (10 cm) flat targets of 50% diffuse reflectance distributed throughout the illuminated field. It should be noted that this technique was derived from experimental observations using stationary objects and observers. VL is defined as the ratio of the target s (photopic) luminance contrast to its threshold luminance contrast at the same background luminance. The threshold contrast is the contrast at which the object is assessed as just becoming visible. VL L L where: L t is the luminance of the object, L b is the luminance of the background, L th is the threshold contrast at the adaptation condition in question. t b = ( 2-1 ) Lth The CIE draft suggests that to provide good visibility, lighting design based on STV should require a smaller number of luminaires and less energy than design based on the minimum illuminance or luminance concepts. A method based on the STV concept is included in the recent American National Standard for Roadway Lighting (ANSI/IESNA, 2000), alongside the traditional methods, so that experience can be gained in its use. Visibility Level (VL) design parameters are discussed in Chapter 4 of the CIE draft document. A need to standardise the size shape and reflectance of the targets, position of observer and target on a grid are highlighted. It is assumed that targets are seen entirely against a road surface. This is facilitated if the angle that they subtend at the observer is limited to 10 min of arc. Observations should be characterised by observer age, which is related to visual performance, and time available for TRL Limited 29 PPR043

40 observation, which is in turn related to stopping time. Account should also be taken of oncoming headlamps. Two methods of calculation are currently used: Adrian s method is based on the VL of flat targets, which is described in CIE (CIE, 1995) and is adopted by the Illuminating Engineering Society of North America (IESNA). Targets are viewed from a constant m ahead of the observer, who s eye height is 1.45 m, giving a constant geometry and target size, but a large number of observer positions. the French method using spherical targets after Lecocq (1999), viewed from a single observer position. In Adrian s method the average of target visibilities over a grid must be higher than the minimum allowed. This averaging can result in some targets not being seen at all. The grid is fixed in size, covering the road between two luminaire positions. Together with the fixed distance and angle criteria, this results in an equal number of observer and target positions. In the French method, the minimum VL anywhere on the grid must exceed a threshold VL of 7. The nearest edge of the grid is set at a luminaire position, 60 m from the observer. The position of far edge of the grid depends on design traffic speed. This is because effective target observation time is inversely proportional to design speed. Flat STV targets are typically 18 cm square, with 20% diffuse reflectance. Spherical targets of a similar reflectance and 20 cm in diameter have a similar apparent surface area. It should be noted that some of the light reflected from the target has first been reflected from the road surface. The Adrian method includes an age factor, representing the falling of visual abilities with age. 2.6 The driving task Visual factors affecting the driving task The task of driving in lit streets involves detection of potential hazards, which may be stationary or moving, in sufficient time to take avoiding action. Because of the limited foveal field in which detail is seen, the eye is constantly searching for hazards. The peripheral field is known to provide much lower visual acuity than the central field, but is very sensitive to changes signifying movement. The relative importance of central and peripheral visual fields is not known, but it is likely that many potential hazards are first detected in the peripheral field, and the eye then moves to bring the object on-axis. The visual performance of the central and peripheral fields is known to differ considerably in acuity and spectral response, so it appears to be important that the relative performance of both should be measured, under the different types of lighting. This is likely to require different designs of experiment. The following factors have all been found to affect the road user s visual abilities at low lighting levels. Most are relevant to all road users, while some are only relevant to drivers of vehicles. These are: Adaptation level Illuminance distribution Illuminance uniformity Lamp spectrum Road surface reflectance TRL Limited 30 PPR043

41 Target object size Target object contrast with background Target object colour and colour contrast with background Target object movement Streetlamp glare Opposing headlamp glare Windscreen light transmission (affected by windscreen angle) Windscreen cut-off angle (from driver s view point) Windscreen cleanliness Road user age In designing experiments to compare the effectiveness of conventional HPS lighting with the prototype CMH lighting, it was necessary to decide which of this large number of possible variables can be held constant and the appropriate ranges for those remaining TRL/ICE experiments Experiments performed for TRL (Cooper and Diver, 2000) compared target detection rates for a range of observers under LPS, HPS and MH lamps, using a 1:20 scale model of a lit road and a 200 millisecond observation time. Variables investigated were: Road surface luminance Lamp type Observer age Glare Background. Fuller details are given in Appendix A. The main conclusions of this work were: Lamp colour there were significantly more target observation errors under the MH lamps than either of the sodium types at levels below 1.5 cd/m 2. Road surface luminance increasing the road surface luminance from 0.5 cd/m 2 to 2.5 cd/m 2 reduced the error rate by only 16%. Glare the errors made by all ages of subject more than doubled when the glare source was introduced. Increasing the road surface luminance did not appear to mitigate this effect. Observer age the older subjects made 1.42 times as many errors and 2.6 times as many false alarms as the younger subjects. As there was nothing to prevent observers using their central visual field in this work it is possible that the lamp colour result is primarily a function of this area of vision. Other work has suggested that MH light is beneficial, particularly in the peripheral visual field. Work to determine the effectiveness of new types of lighting should therefore include separate experiments to investigate central and peripheral vision. TRL Limited 31 PPR043

42 2.7 Implications for the design of laboratory and field experiments The objective of the experimentation was to determine whether the prototype lighting makes it possible to reduce the level of illumination of the highway without any detrimental effects on road safety and road user comfort. The benchmark for the comparison was a 150W high-pressure sodium lamp. This was chosen because it was considered the most commonly used in urban lighting schemes. A prototype lamp CMH of 150W was the subject of the project but development of 250W and 400W CMH lamps to give similar characteristics was also envisaged. It is clear from the foregoing that the performances of the central and peripheral visual fields differ at any level of illumination. Although the literature reviewed gives little indication of the relative importance of these fields to the road user, it is clear that they are both essential. The central field possesses far greater visual acuity and the best colour vision. This is needed for the identification and recognition of detected objects. However this field is small and must be directed to points of interest by eye and even head movement, which is a relatively slow process. The peripheral field is large and although it has inferior acuity and recognition performance, it is very sensitive to movement and change. It is essential for the detection of objects that may prove to be a hazard if they are close to or approaching the road user s path. Lacking more precise information, it will be assumed that the central and peripheral fields are of equal importance to the road user. The literature indicates that possible beneficial effects on vision of lamps emitting a broader spectrum with more light energy emitted at the shorter wavelengths may only be apparent in the peripheral visual field. If this is to allow illumination to be reduced it will be important to ensure that the identification of potential hazards by the central field is not compromised. It was therefore necessary to perform experiments that separated the effects, on the central and peripheral fields, of changes in lighting spectrum and illumination level. It has not always been clear in previous research which part of the visual field was being tested. Detection times or detection distances of targets of various types were considered the most appropriate parameters to the road user s task. These have been measured in a number of ways. The experiments reviewed can be categorised as shown in Table It can be seen that few of the experiments were able to separate the effects of the variables investigated on the central and peripheral fields. Only one experiment attempted to include the effects of relative movement of the observer and the scene on reaction time. In all others, the experiment was static and dynamic effects on observation time were simulated by restricting viewing per observation to about 0.2 seconds. Table Summary of types of test in the literature Type of test LABORATORY TESTING Presentation of simple visual patterns, for central or peripheral vision A physical scale-model of a road Computer display using patterns Measured Central (c) Peripheral (p) (separately) Relative movement (observer target) Main variables Time to detect a target c, p None Lamp type, illuminance, observer Time to detect a target both None Lamp type, illuminance, observer, target position Time to detect a target both None Target contrast, colour, luminance, observer TRL Limited 32 PPR043

43 Computer display using virtual models FIELD TESTING USING FULL-SCALE LIT ROADS Detection of simple targets in the road Detection of peripheral targets Dynamic task accuracy both Yes Light spectrum, task, observer Subjective assessment of visibility both None Lamp type, illuminance, target position, observer Time to detect change p None Lamp type, illuminance, target position, opposing headlamps Instruments The instruments, used in the experimental work, were a spectroradiometric illuminance meter/luminance spotmeter and a photometric digital camera. Details are given in Appendix G. 2.8 Conclusions A number of important conclusions can be drawn from the literature review: The human visual system The adaptation process is complex and not fully understood. The change in pupil size has only a minor role in adaptation. Poor visual acuity (sharpness of an object) is found in the peripheral field because of the relatively few cones compared to rods. Acuity is reduced by chromatic aberration of the lens of the eye. The opening of the iris has a detrimental effect on visual acuity by increasing the spherical aberration and also by reducing the effective depth of the focus of the eye. These are the main causes of the phenomenon known as night myopia. Published values for peak sensitivity vary at least from 440nm to 498nm for the S-cones (short wavelength), 535 to 545nm for the M-cones (medium wavelength) and 564nm to 580nm for the L-cones (long wavelength). In the conventional model of vision, luminance information is derived from the summation of the signals from all three of these channels, while chrominance information is derived from the differences between them. The brain uses these chrominance signals to compute the sensation of colour. Other published work gives evidence of four types of receptor in the human eye, which have peak sensitivities at 342nm, 437nm, 532nm and 625nm. In particular, the monochromatic rods and colour cones model is shown to be incorrect. The CIE curves are greatly smoothed versions of the true response, based on limited measurement capabilities in the 1930's. The visual system adapts to colour. Flicker-photometry was developed to overcome the effect of this on side-by-side brightness matching experiments. The macula, which overlays the fovea and foveola, appears to be yellow but its function is not fully understood. In the mesopic zone, lighting that stimulates the short and medium wavelength receptors is likely to be more effective than the longer wavelengths. The shape of the visual response curve changes throughout the mesopic zone in a complex and poorly defined manner. As luminance decreases through the mesopic zone, there is a loss of colour sensitivity. TRL Limited 33 PPR043

44 Photopic luminance is not a reliable guide to the performance of the visual system particularly at the levels used in street lighting and does not correlate well with perceived brightness. The road user s task Brightness matching has not been successful in defining the mesopic vision model and is not necessarily the most important factor to road users under street lighting. Visual reaction time is considered to affect road user performance more than visual acuity. Detection times or detection distances of objects are considered the most appropriate parameters to the road user s task. Object detection performance of the peripheral field is also important. Road users are susceptible to rapid changes of illumination if light levels change faster than can be accommodated by the adaptation mechanism (adjustment to light level). Reaction times are not significantly affected by opposing headlamp glare. As luminance levels increase glare is increased. Older people take longer to recover from exposure to glare. Discomfort glare is task dependent and may be more important overall than disability glare, as it is more difficult to prevent at street lighting levels. Satisfactory performance of the road user s task requires both central vision and peripheral vision. Central vision is little affected by changes in the SPD of the lighting. However, the peripheral field, which is important for the detection of hazards, is very sensitive to the SPD of the lighting. None of the British and European standards attempt to account for spectral effects and all are based on photopic lumen measurements. These are used in urban mixed pedestrian and vehicle situations where the luminance of the background is less certain. TRL Limited 34 PPR043

45 3 Laboratory experiments measuring reflectance on three pavement surfaces using CMH and HPS lamps 3.1 Objectives Discussions between Thorn Europhane, GE Lighting (GEL) and TRL determined that the main objectives of the laboratory work should be the measurement of the reflectance parameters of the surfaces in the trial and the investigation of any difference in reflectance of the road surfaces due to the spectral differences between a standard high pressure sodium (HPS) lamp and a prototype ceramic metal halide (CMH) lamp. 3.2 Test samples The road surfacing samples measured were: A core from the site of the future trial on the TRL Small Road System (SRS). A core from the surface of the Rue Bèrchere in Albi. A sample of Stone Mastic Asphalt (SMA) which had been subjected to the equivalent of two years trafficking on the TRL road machine. Figure 3.1 SRS sample (left), Albi samples (centre) and SMA sample (right) The SRS has a surface-dressing of 10mm chippings, which is relatively unworn because the system has very light traffic (Figure 3.1, left frame). The core tested from the Rue Bèrchere in Albi is shown in Figure 3.1. This was one of two cores which were taken from opposite sides of the carriageway, 1m from the kerb. The cores are of a densely graded asphalt mix of maximum aggregate size 10mm, and are likely to be a semi coarse bituminous concrete (BBSG). The SMA road machine sample is shown in the right frame of Figure 3.1. This is an example of the semi-porous thin surfacing systems which are now specified for most surfacing of trunk roads and motorways in the UK. 3.3 Road surface reflectance measurements in lighting design In the design of street lighting, tables of reduced luminance coefficients (r tables) define the reflectance characteristics of road surfaces. At each point on the road surface, the luminance seen at a standard viewing angle can be obtained by multiplying the r-coefficient for that point by the luminous intensity of the luminaire in the direction of that point. The form and method of measurement of the data in these tables are defined by the International Commission on Illumination (CIE/PIARC, 1983). The tabulated coefficients are distributed over a polar grid on the road surface centred in plan vertically below a luminaire. The coefficients define the reflectances of the road surface seen from an TRL Limited 35 PPR043

46 eye height of 1.5m at a standard viewing angle of 1 below the horizontal. This model is considered to be sufficiently accurate over a range of viewing distances from 60m to 170m along the road. The tables relate to a roughly rectangular area of the road with a width of three times the height of the luminaire. The area extends to twelve times this height in front of and four times this height behind the luminaire as seen from the viewing position. The full r table contains 396 coefficients and is therefore difficult to measure without automated equipment. Few reflectometers are available which can perform this high-precision laboratory work. As a result, sets of standard r-tables published by the CIE (CIE/PIARC, 1983) are generally used in lighting design. Each set of tables complies with one of three classification systems, the older R and N systems, which each contain four classes and the newer C system, which contains only two classes. In each system, the class boundaries are determined by the value of the specular factor S1, which is roughly equivalent to shininess. S1 is the ratio of the r-value at a point two column-heights directly towards the observer with the r-value directly beneath the luminaire. Each table contains r-values to be used for a specified range of S1 values. The normalised luminance coefficient Q 0, which is equivalent to lightness, is the appropriately weighted sum of all 396 r- values in the full table. The Q 0 values for which the tables are compiled have been chosen to represent the average value found in each class. If the Q 0 of the surface to be classified is different from that of the table selected then all the r-table values should be proportionally adjusted. The significance of the Q 0 value is that the figure is inversely proportional to the luminous flux required from the lamps (Van Bommel and de Boer, 1980). Hence, an increase in the value of Q 0 is desirable in terms of lighting efficiency, as long as luminance uniformity is not compromised. It was decided that as the older R system has a finer differentiation across the range, this system would be used for the classification of the test surfaces (see Table 3.1). Table 3.1 Standard R classification system (CIE/PIARC, 1983) Class r-table S1 of r-table S1 limits Normalised Q 0 Typical surfaces RI R S1 < Concrete RII R ] S1 < Bituminous RIII R ] S1 < Bituminous RIV R S Bituminous Because it is not practical to measure full r-tables for every road surface, the standard tables are commonly used and approximate Q 0 values are estimated based on the generic type of the surface. Field instruments for the measurement of the diffuse reflectance coefficient Q d, exist, but Schreuder (1998) states that it is a bad predictor of Q 0. However Sorensen (2003) has provided the following information: Q d values are generally lower than Q 0 values, to a degree depending on the specularity of the road surface. For standard r-tables R1, R2, R3 and R4, this is by respectively 13%, 19%, 29% and 35%. The standard r-tables apply to dry road-surface conditions only. Measured reflectance parameters will also depend on the amount of wear that the test specimens have undergone. The CIE recommends that samples should be in a state representative of most of their life and should therefore not be less than one year old. The reflection characteristics are likely to vary with lateral position (outside wheeltracks, in wheeltrack, between wheeltracks), weather conditions and contamination, particularly by oil from vehicles. For the current wet condition measurements, a different set of classes was used. Fredriksen and Sorensen (1976) have devised four road-surface classes (Table 3.2) valid for typical wet conditions on TRL Limited 36 PPR043

47 road surfaces measured in Norway. The classification is based on the corrected specular factor S1'. This is related to S1 as follows: For S1 > 1 log( S1' / 0.147) log( S1/ 0.147) = 1 / Q 0 and for S1 ] 1 S1' = S1 where the S1 and Q 0 values are those for wet conditions. The class limits are then defined using this corrected specular factor (see Table 3.2). Table 3.2 Wet road surface class limits Class S1' limit W1 S1 '< 9.6 W2 9.6 ] S1' < 26.5 W ] S1' < 73 W4 S1' Experimental design The experiments were designed to determine any differences between the reflectances of road surfaces under a prototype CMH lamp, provided by GE Lighting in August 2003, and the standard GE Lucalox HPS lamp chosen as a reference. Experimental apparatus was designed and built which permitted the measurement of 28 r-table coefficients. The angles for these measurements were chosen to be well distributed over the r-table, as shown in Table 3.3. Positions on the rig were labelled 1 to 8 around the vertical axis (azimuth angle, ^) and a to g around a vertical quadrant (elevation angle, _). These labels are shown in Table Experimental apparatus The rig comprised a quadrant beam, pivoted at the top about a vertical axis and movable at its foot which was located on a semi-circular track fixed to the floor, as shown in Figure 3.2 and Figure 3.3. The lamp holder and reflector were mounted on this quadrant and could be moved to each of the marked positions. The combination of these positions (_) with the different positions of the quadrant on the track (^) provided the required 28 illumination angles. The specimen was precisely located in a steel tray on three footscrews, as illustrated in Figure 3.4, which provided levelling and height adjustment. The height of the centre of the specimen was set by means of a fixed vertical reference probe mounted on a removable steel bridge. A spirit level was placed on the sample surface and the footscrews adjusted until a level position was achieved. The bridge and level were removed prior to measurements being made. TRL Limited 37 PPR043

48 Table 3.3 Chosen r-table coefficients for measurement ^ (degrees) _ (degrees) Tan _ g f - 4f - 2f e - 6e 5e 4e 3e 2e 1e d - 6d 5d 4d 3d 2d 1d c 7c 6c 5c b 7b 6b a 7a 6a Figure 3.2 Schematic of reflectance rig TRL Limited 38 PPR043

49 Semi-circular track Quadrant track Lamp holder Camera Sample Figure 3.3 Reflectance rig Sample Base plate Footscrews Figure 3.4 Sample Leveller The light source was either a CMH lamp or an HPS lamp mounted in an E40 lampholder and a reflector provided by Thorn Europhane. This was masked by a sheet metal tube attached to the front face of the reflector, to provide illumination of the specimen with minimal overspill (see Figure 3.5). TRL Limited 39 PPR043

50 Figure 3.5 Lamp masking tube 3.6 Instruments Photometric camera The photometric camera used was an LMK Mobile system from Techno-Team GmbH. This is based on a Rollei D30flex professional digital camera, together with modified firmware and the LMK 2000 acquisition and analysis software running on a host computer. The camera can be used independently of the computer up to the limit of its onboard storage. The Charged Coupled Device (CCD) used has a pixel size of 1280 (h) x 1024 (v), but for luminance measurements groups of four pixels are combined, which reduces the photometric luminance resolution to 640 x 512. At its maximum focal length, the zoom lens gives a 26º horizontal field of view, corresponding to 46.2m at 100m distance. This results in a luminance resolution of 2.4 arc-minutes, both vertically and horizontally. The system uses exposure bracketing to overcome the dynamic range limitations of the CCD, which results in a reported dynamic range of 1:50,000. The resolution and minimum measurable (photopic) luminance are both 0.01cd/m 2. Each group of four pixels in the CCD comprises two with a green, one with a red and one with a blue filter. The signals from these are manipulated to give a weighted output. A CIE V (S) weighting is provided, but adjustments to this can be made using the computer software if required Spectroradiometer The Jeti Specbos 1200 spectroradiometer is a compact luminance meter capable of 5nm wavelength resolution over the range 380nm to 779nm. The software provided calculates average radiance and a range of photopic parameters, including luminance (illuminance with a cosine attachment), colour rendering indices, correlated colour temperature and CIE colour coordinates. The device is powered from the host computer via the USB bus and is controlled remotely via software on the PC. This allows operation without disturbance of the measuring instrument. The acceptance angle is 2 degrees and the measuring area can be marked on the target by a projected red laser circle and centre spot. TRL Limited 40 PPR043

51 3.6.3 Powermeter A Metrahit 29s digital multimeter/powermeter was used in conjunction with a dedicated clip-on current transformer to determine the power drawn by the luminaire. This was used to confirm the correct function of the ballast by indicating that the expected power was being drawn. 3.7 Experimental Method Measurement of illuminance The illuminance at the centre of the sample was measured using the Specbos 1200 spectroradiometer and a standard diffuse 98 percent reflectance plate. The specimen was removed and the diffuse reflectance plate put in its place. The Specbos was then used to measure the luminance of the plate. For this purpose, the instrument was mounted on a tripod with its axis vertical, centred on the reflective tile. For the top position of the lamp, giving vertical illumination, the instrument was moved to one side until it did not cast a shadow on the plate, while ensuring that the measuring area was still fully on the plate. These measurements were used to calculate the photopic illuminance and the radiance spectrum for each of the 28 illumination angles and for each of the two lamps. These data were used subsequently to calculate the reduced luminance coefficients of the road surface samples and for comparison with spectra of the reflected light from the samples Measurement of sample luminance The reflected light was measured using the Rollei D30 photometric camera, at such a height as to realise the specified observation angle of 1 at the centre of the specimen. The camera was attached to a pedestal which was solidly bolted to the floor. The camera remained undisturbed for the duration of the measurements. The camera lens was shielded from direct light from the luminaire by a matt-black mask (see Figure 3.6). Use of the camera allowed the full surface of the specimen to be mapped at high resolution. Figure 3.6 Camera masking The Specbos was also used to measure the luminance of the surface samples, to complement the camera measurements. The instrument was mounted on a small tripod, as close as possible to the sample without casting a shadow on it for any position of the lamp. It was found that to keep the measuring area fully on the sample, the angle between the instrument axis and the sample surface could be no lower than 3.5. After analysis of some initial data, it was realised that unwanted direct light from the lamp was causing lens flare and being measured when the lamp was in front of the instrument. Hence a mask was installed to intercept all incident light from sources other than the TRL Limited 41 PPR043

52 sample surface (Figure 3.7). Figure 3.7 Spectrometer masking Measurements were carried out on the following samples: A positive textured surface from TRL a surface dressed sample from the TRL small road system. A negative textured sample from TRL stone mastic asphalt (SMA). A sample from the pilot site in Albi, France believed to be semi-coarse or high modulus bituminous concrete. The SMA sample had been trafficked for the equivalent of two years on the TRL road machine (see Figure 3.8). Under normal operation, the circular driven table rotates and two standard car tyres (size 195/70/14), mounted vertically over the table, run over surface samples in the test bays. The tyre assembly is loaded at 5kN to represent normal traffic loading. The table holds asphalt samples with dimensions of 305mm x 305mm x 50mm. The Albi sample was a 200mm-diameter core taken from the pavement of the Rue Berchère on the trial site in Albi. The initial objective was to test each material sample in both dry and wet surface conditions. With the two light sources, three surfaces, two conditions and 28 measurement angles; this gave a total of 336 measurements. In practice 308 measurements were made because the SMA sample was not tested in the wet condition. This was because the surface did not appear to wet in the same way as a road surface in use, the water ponding and forming beads. This may have been a result of lack of weathering from outdoor exposure, or an excess of binder material on the surface. TRL Limited 42 PPR043

53 Figure 3.8 TRL road machine Wet condition In practice, road conditions vary between completely flooded and dry. For a flooded road, the light sources are mirrored in the water surface, so the reflectance of the road surface itself is not relevant. The CIE reference (CIE/PIARC, 1983) states that the luminance coefficient concept is valid for wet but not flooded surfaces. Van Bommel and de Boer (1980) state that, in comparison with the dry condition, the average luminance of a road surface increases in wet conditions and the uniformity of the luminance decreases. The decrease in uniformity is seen as the bigger problem. The CIE has defined a standardised wet surface condition for laboratory specimens as the state 30 minutes after the surface has been uniformly sprayed, at the rate of 5mm of rain/hour, in a draughtfree room at a temperature of 25 C and a relative humidity of 50%. It is possible to construct r-tables for wet surfaces, but these will be specific to the particular wet conditions as well as the type of surface. Frederikson and Sorenson (1976) created wet condition r-tables (Class W1 to W4 ) for four classes of road surface common in Scandinavia at that time. These tables are presented in Appendix B of Van Bommel and de Boer (1980). It was considered that the CIE method produced a ponded surface which was not representative of typical wet conditions. It was therefore necessary to determine a method of creating a repeatable wet condition for the laboratory tests that was sustainable over a large number of readings. In order to maintain the sample in the wet condition, a distilled water spray was used between each reading. Surplus water was wiped off with a paper towel to eliminate ponding effects (see Figure 3.9). While the position of the lamp was adjusted the surface of the specimen was covered with polystyrene. This eliminated any effect of radiant heat from the lamp and reduced evaporation of water from the specimen. TRL Limited 43 PPR043

54 Figure 3.9 Sample wetting procedure 3.8 Results Road surface illuminance measurements Figure 3.10 and Figure 3.11 show the illuminance of the CMH and HPS lamps on the perfectly diffusing surface, with respect to the angle of incidence of the illumination. The surface was mounted in the same position as the road surface samples. The graphs confirm that the illuminance on the surface, calculated from its measured luminance, closely follows the cosine law to the largest incidence used. TRL Limited 44 PPR043

55 CMH Illuminance against cos(gamma) Illuminance (lux) Cos (gamma) Figure 3.10 CMH illuminance against incident angle standard diffusing surface HPS Illuminance against cos(gamma) Illuminance (lux) Cos (gamma) Figure 3.11 HPS illuminance against incident angle standard diffusing surface TRL Limited 45 PPR043

56 3.8.2 Photometric camera The output from the camera analysis software provided a luminance image with the sample surface near the centre. Ideally, the luminance at a point in the centre of the sample is required. However, in practice it is necessary to average the luminance of a large enough area to encompass a representative area of the surface texture. A small target area of width 10 pixels (N12.4mm at the centre of the sample) and height 3 pixels (N93mm in the target plane) was superimposed on the image to ensure consistency between samples. The target area avoided the edges of the sample surface to eliminate any error from pixels which were only partly on the surface. A typical photograph is shown, with the measuring area marked by a small rectangle, in Figure The luminance was averaged by the software over this area only, for each of the 28 angles of illumination Spectroradiometer The results from the spectroradiometer were saved directly to a spreadsheet, which gave the radiance at each wavelength from 380nm to 779nm. For each surface and lamp this enabled the comparison of the spectrum of the reflected light at each illumination angle with the spectrum of the incident light at the same angle, measured as described in Section 3.7. (Note that the spectrum of the incident light was not expected to change with angle.) Figure 3.12 Luminance image screenshot TRL Limited 46 PPR043

57 3.8.4 Surface dressed small road system surface In order to calculate values for S1 (and S1' for the wet observations), and estimate Q 0, partial r-tables were constructed for each surface as shown in Table 3.3. For each table, the value of S1 was derived from the ratio of two of the r-coefficients as described in Section 3.3. S1 was then used to place the data in the correct CIE class. The average of the ratio of each of the 28 measured r-values with the corresponding r-value in the standard table, for this S1, was calculated. This factor was then used to derive the Q 0 for the measured surface from that for the tabulated class. All the partial r-tables for the LMK camera data are presented in Appendix F, together with the appropriate extracts from the CIE standard tables (dry observations) and the Frederiksen and Sorensen (1976) tables (wet observations). Table 3.4 shows values for S1, S1' and Q 0 calculated in this way, using both the Specbos spectroradiometer and the LMK mobile camera and analysis software. Table 3.4 Small road system surface Small road system CIE surface S1 S1' Qo Class Camera, dry, HPS NA * R1 1 degree CMH NA R1 Camera, wet, HPS NA R3 1 degree CMH NA R3 Specbos, dry, HPS NA R1 3.5 degrees CMH NA R1 Specbos, wet, HPS W1 3.5 degrees CMH W1 * NA = not applicable Figure 3.13 shows the spectrum of the reflected light at each lamp angle, normalised to an arbitrary value of 1.0 at 650nm, so that the spectra can be compared. The normalisation wavelength of 650nm was chosen because it was observed that at longer wavelengths the spectra were a good match, but at shorter wavelengths that there was some divergence. The data were obtained from the spectroradiometer and are for the CMH lamp with a dry sample. TRL Limited 47 PPR043

58 SRS - dry - CMH lamp - reflected light Normalised radiance Wavelength (nm) 8g (r) 8e (r) 8d (r) 8c (r) 8b (r) 8a (r) 7a (r) 7b (r) 7c (r) 6f (r) 6e (r) 6d (r) 6c (r) 6b (r) 6a (r) 5c (r) 5d (r) 5e (r) 4f (r) 4e (r) 4d (r) 3d (r) 3e (r) 2e (r) 2f (r) 2d (r) 1d (r) 1e (r) Figure 3.13 Normalised wavelength profiles for CMH lamp, SRS surface dry Figure 3.14 represents the equivalent data for the HPS lamp. Normalised radiance SRS - dry - HPS lamp - reflected light 8g 8e 8d 8c 8b 8a 7a 7b 7c 6f 6e 6d 6c 6b 6a 5c 5d 5e 4f 4e 4d 3d 3e 2e 2f 2d 1d 1e Wavelength (nm) Figure 3.14 Normalised wavelength profiles for HPS lamp, SRS surface dry TRL Limited 48 PPR043

59 Figure 3.15 shows the data obtained from the spectroradiometer for the CMH lamp with a wet sample. SRS - wet - CMH lamp - reflected light Normalised radiance Wavelength (nm) 8g 8e 8d 8c 8b 8a 7a 7b 7c 6f 6e 6d 6c 6b 6a 5c 5d 5e 4f 4e 4d 3d 3e 2f 2e 2d 1d 1e Figure 3.15 Normalised wavelength profiles for CMH lamp, SRS surface wet The graph in Figure 3.16 represents the equivalent data for the HPS lamp. Normalised radiance SRS - wet - HPS lamp - reflected light Wavelength (nm) 8g 8e 8d 8c 8b 8a 7a 7b 7c 6f 6e 6d 6c 6b 6a 5c 5d 5e 4f 4e 4d 3d 3e 2f 2e 2d 1d 1e Figure 3.16 Normalised wavelength profiles for HPS lamp, SRS surface wet TRL Limited 49 PPR043

60 3.8.5 Sample surface from the Rue Berchère Albi, France Table 3.5 shows values for S1, S1' and Q 0 calculated from the partial r-tables obtained from the experiments, using both the Specbos spectroradiometer and the LMK mobile camera and analysis software. Albi surface Table 3.5 Albi surface CIE S1 S1' Qo Class Camera, dry, HPS NA * R3 1 degree CMH NA R3 Camera, wet, HPS W1 1 degree CMH W1 Specbos, dry, HPS NA R3 3.5 degrees CMH NA R3 Specbos, wet, HPS W4 3.5 degrees CMH W4 NA* = Not applicable For comparison, Thorn Europhane commissioned the Laboratoire Regional de Ponts et Chaussées at Cleremont Ferrand (LRPC) to undertake measurement of a full set of r-table data, on the same core, using the standard CIE incandescent light source. The observation angle was 1.3. The resulting Q 0 value was 0.063, which gave in a classification of R3. This Q 0 is a little higher than the TRL value reported here (0.05) for the CIE specified observation angle of 1, but identical to the Q 0 value obtained at 3.5. As the measurement obtained will vary rapidly with observation angle, the agreement between the Q 0 values obtained at TRL and at LRPC is therefore good considering the differences in the test methods used. TRL Limited 50 PPR043

61 Figure 3.17 shows the spectrum of the reflected light at each lamp angle, normalised to a value of 1.0 at 650nm, so that the spectra can be compared. The data were obtained from the spectroradiometer and are for the CMH lamp with a dry sample. Normalised radiance Albi - dry - CMH lamp - reflected light Wavelength (nm) 8g (r) 8e (r) 8d (r) 8c (r) 8b (r) 8a (r) 7a (r) 7b (r) 7c (r) 6f (r) 6e (r) 6d (r) 6c (r) 6b (r) 6a (r) 5c (r) 5d (r) 5e (r) 4f (r) 4e (r) 4d (r) 3d (r) 3e (r) 2e (r) 2f (r) 2d (r) 1d (r) 1e (r) Figure 3.17 Normalised wavelength profiles for CMH lamp, Albi surface dry Figure 3.18 represents the equivalent data for the HPS lamp. Normalised radiance Albi - dry - HPS lamp - reflected light Wavelength (nm) 8g 8e 8d 8c 8b 8a 7a 7b 7c 6f 6e 6d 6c 6b 6a 5c 5d 5e 4f 4e 4d 3d 3e 2e 2f 2d 1d 1e Figure 3.18 Normalised wavelength profiles for HPS lamp, Albi surface dry TRL Limited 51 PPR043

62 Figure 3.19 shows the data obtained from the spectroradiometer for the CMH lamp with a wet sample. Albi - Wet - CMH lamp - reflected light Normalised radiance Wavelength (nm) 8g 8e 8d 8c 8b 8a 7a 7b 7c 6f 6e 6d 6c 6b 6a 5c 5d 5e 4f 4e 4d 3d 3e 2f 2e 2d 1d 1e Figure 3.19 Normalised wavelength profiles for CMH lamp, Albi surface wet The graph in Figure 3.20 represents the equivalent data for the HPS lamp. Normalised radiance Albi - wet - HPS lamp - reflected light Wavelength (nm) 8g 8e 8d 8c 8b 8a 7a 7b 7c 6f 6e 6d 6c 6b 6a 5c 5d 5e 4f 4e 4d 3d 3e 2f 2e 2d 1d 1e Figure 3.20 Normalised wavelength profiles for HPS lamp, Albi surface wet TRL Limited 52 PPR043

63 3.8.6 Stone mastic asphalt road machine surface Table 3.6 shows values for S1, S1' and Q 0 calculated from the partial r-tables obtained from the experiments, using both the Specbos spectroradiometer and the LMK mobile camera and analysis software. SMA surface Table 3.6 SMA surface CIE S1 S1' Qo Class Camera, dry, HPS NA * R3 1 degree CMH NA R3 Camera, wet, HPS NA NA NA NA 1 degree CMH NA NA NA NA Specbos, dry, HPS NA R4 3.5 degrees CMH NA R4 Specbos, wet, HPS NA NA NA NA 3.5 degrees CMH NA NA NA NA * NA = not applicable TRL Limited 53 PPR043

64 Figure 3.21 shows the spectrum of the reflected light at each lamp angle, normalised to a value of 1.0 at 650nm, so that the spectra can be compared. The data were obtained from the spectroradiometer and are for the CMH lamp with a dry sample. Normalised radiance SMA - dry - CMH lamp - reflected light Wavelength (nm) 8g (r) 8e (r) 8d (r) 8c (r) 8b (r) 8a (r) 7a (r) 7b (r) 7c (r) 6f (r) 6e (r) 6d (r) 6c (r) 6b (r) 6a (r) 5c (r) 5d (r) 5e (r) 4f (r) 4e (r) 4d (r) 3d (r) 3e (r) 2e (r) 2f (r) 2d (r) 1d (r) 1e (r) Figure 3.21 Normalised wavelength profiles for CMH lamp, SMA surface dry The graph in Figure 3.22 represents the equivalent data for the HPS lamp. Normalised radiance SMA - dry - HPS lamp - reflected light Wavelength (nm) 8g 8e 8d 8c 8b 8a 7a 7b 7c 6f 6e 6d 6c 6b 6a 5c 5d 5e 4f 4e 4d 3d 3e 2e 2f 2d 1d 1e Figure 3.22 Normalised wavelength profiles for HPS lamp, SMA surface dry TRL Limited 54 PPR043

65 3.8.7 Comparison of incident and reflected spectra Figure 3.23 to Figure 3.26 compare the incident and reflected spectra for selected angles of measurement, for the dry Albi surface and the CMH and HPS lamps. The elevation angle _ at position 8g is 0 and at position 1e _ is Albi core - dry - CMH lamp - incident and 3.5 degree reflected spectra Normalised radiance g Incident 8g reflected Wavelength (nm) Figure 3.23 Albi surface, dry, under CMH illumination, position 8g Albi core - dry - CMH lamp - incident and 3.5 degree reflected spectra Normalised radiance e Incident 1e reflected Wavelength (nm) Figure 3.24 Albi surface, dry, under CMH illumination, position 1e (note different vertical scale) TRL Limited 55 PPR043

66 Albi core - dry - HPS lamp - incident and 3.5 degree reflected spectra Normalised radiance g Incident 8g Reflected Wavelength (nm) Figure 3.25 Albi surface, dry, under HPS illumination, position 8g Albi core - dry - HPS lamp - incident and 3.5 degree reflected spectra Normalised radiance e Incident 1e Reflected Wavelength (nm) Figure 3.26 Albi surface, dry, under HPS illumination, position 1e TRL Limited 56 PPR043

67 3.9 Discussion of results It can be seen from the incident light spectra in Figure 3.27 that with the CMH lamp at position 8g, which is directly above the sample, the measured spectrum from the lamp differed significantly from that measured in other positions. This did not occur with the HPS (Figure 3.28), demonstrating that this effect is a property of the CMH lamp. GE Lighting has confirmed that the spectrum of the lamp changes with the orientation of its axis and takes several hours to stabilise after a change. This effect was not appreciated during the design of the experiment and the lamp was mounted so that in position 8g the lamp was horizontal, and as the elevation angle ^ varied, the lamp axis rotated towards vertical. Therefore, the spectrum of the CMH lamp will never have been at equilibrium during the measurements as its position was continually being changed. Figure 3.27 shows the measured incident light spectra for all 28 positions. It is clear from this that there is a large change in spectrum between positions even without allowing time for equilibration. It is also clear from the figure that the largest part of this change occurred between the top position 8g, where the lamp is horizontal and the four positions f where the lamp axis is tilted Figure 3.23 and Figure 3.24 show that despite the foregoing, good agreement was obtained between the incident and reflected spectra for some measurements, including position g. This demonstrates that the larger differences observed between incident and reflected spectra for other measurements are most likely to have been due entirely to the orientation of the lamp being changed, and that different times had elapsed since the change for the incident and reflected light measurements. Peaks in the spectrum occurred at approximately 410nm, 537nm and 596nm for the CMH lamp, with peaks at 569nm, 582nm and 599nm for the HPS lamp. Comparing the profile of the spectra for incident light with those for reflected light, neither of the lamp spectra appeared to experience a significant shift on any of the sample surfaces. However, it can be seen from Figures 3.13, 3.17 and 3.19 that in the lower part of the spectrum, below 530nm, there was more spread between the spectra plotted for the wet condition, than for the dry, particularly for the SRS surface. However, comparison with the incident light spectra in Figure 3.27 suggests that all these variations were caused by the lamp orientation effect. TRL Limited 57 PPR043

68 CMH lamp - std target - incident light (normalised) Normalised radiance Wavelength (nm) 8G 8E 8D 8C 8B 8A 7A 7B 7C 6F 6E 6D 6C 6B 6A 5C 5D 5E 4F 4E 4D 3D 3E 2F 2E 2D 1D 1E Figure 3.27 All incident light spectra for CMH lamp HPS lamp - std target - incident light (normalised) Normalised radiance Wavelength (nm) 8G 8E 8D 8C 8B 8A 7A 7B 7C 6F 6E 6D 6C 6B 6A 5E 5D 5C 4F 4E 4D 3D 3E 2F 2E 2D 1D 1E Figure 3.28 All incident light spectra for HPS lamp At the specified observation angle of 1 the camera data showed that the Albi surface was lighter and more specular than the SRS surface when the surfaces were dry. This is explained by the considerable difference in surface texture, the Albi surface being much smoother than that of the SRS. The SMA TRL Limited 58 PPR043

69 surface has similar properties to the Albi surface. In all cases there was little difference between the Q 0 and S1 values under CMH and HPS light, for any of the surfaces. At this observation angle, the wet samples for both the Albi surface and the SRS surface had a lower overall luminance than that of the dry samples. As expected the specularity increased dramatically for both surfaces when wet. However, for the surface dressed sample the wet S1 value still allowed it to be placed in (dry) Class R3, whereas the more specular Albi surface moved to the (wet) Class W1. The increased specularity when wet leads to a much lower luminance uniformity on the road. When dry, the Q 0 and S1 from the Specbos observations, at the higher observation angle of 3.5, were similar for the Albi and SRS surfaces, but the specularity of the SMA doubled compared with that at the 1 angle. When wet the Specbos data showed dramatic increases in both Q 0 and S1 compared with the camera data. This emphasises the difficulty in obtaining meaningful lighting design data for wet roads, where all parameters are likely to vary greatly with the viewing direction, due to the very high specularity Conclusions The work described has provided measurements of the specular factor S1 and estimates of the luminance, Q 0, for three road surfaces, in both the wet and dry conditions. These figures were provided to Thorn Europhane to assist with the calculation of expected luminances, uniformities and theoretical small target visibilities for the full-scale trial at TRL and the pilot installation in Albi. The experimental data confirms that the CMH, and not the HPS, lamp spectrum changes with orientation of the lamp axis with respect to the vertical. More short wavelength light from the CMH lamp was measured when the lamp axis was vertical. The measurements demonstrate that this effect was responsible for the changes in spectrum seen in this experiment. No differences were observed between incident and reflected spectra which could be attributed to reflection. The spectral data is important because the mesopic effect to be tested at TRL depends on the presence of the shorter wavelengths in the CMH light than in the HPS light. Selective absorption of these wavelengths would reduce their intensity in the reflected light. TRL Limited 59 PPR043

70

71 4 Assessment of visual performance under CMH and HPS lighting systems in a full-scale trial In order to assess the visual performance of observers under both CMH and HPS lighting systems, a pilot-scale installation was commissioned on the TRL small roads system, shown in Figure 4.1 and Figure 4.2. On the basis of previous research in this area, described earlier, it was decided that observers would be asked to perform two separate tasks in the trial. The aim of the first task was to investigate the observers ability, whilst driving at a low constant speed, to detect a small target placed on the road surface in various positions. This required the collection of accurate detection distances for different target, lamp and light output combinations. The task demanded use of the entire visual field, which includes the central field of vision. The visual field is commonly defined as the total area in which objects can be seen in the peripheral vision while the eye is focused on a central point. The second task involved an assessment of an observer s ability to detect a change in luminance of a target positioned in their peripheral field of vision. The requirements of this experiment were to record reaction times under various lighting conditions and peripheral target locations. A description of the lighting installation installed to fulfil these requirements is given in Section 4.1. Further details on the generic and task-specific measurements undertaken using the lighting installation are described in Section Description of lighting installation The trial length consisted of a road surface with a surface-dressing of 10mm chippings, as shown in Figure 4.1, which is part of the TRL Small Road System (SRS). The trial length was approximately 350m long, 9m wide with 2m wide footways on each side and lined on both sides by trees. The length consisted of 100m of unlit road, followed by the lit length of approximately 150m followed by a further unlit length of 100m. This is shown schematically in Figure 4.2. The lighting installation comprised five 9m high lamp columns at a spacing of 30m. This was designed to match the geometry of part of the demonstration installation in the town of Albi, France. Figure 4.1 View of SRS trial pilot scale lighting installation (note: detritus was cleared in advance of all trials) TRL Limited 61 PPR043

72 Figure 4.2 Schematic diagram of the track layout and target positions Figure 4.3 Lighting column and Thorn Decostreet 2 luminaires at centre spacing of 0.9m Each column supported a specially-fabricated bracket, enabling the mounting of two Thorn Decostreet 2 luminaires at a centre spacing of 0.9m as illustrated in Figure 4.3. This enabled the light source to be switched from HPS to CMH and the reverse, as often as required during the trial without the need for physical access. The road was swept prior to the start of the trials and kept clear of fir-cones during the trial. Several low branches near luminaires were cut back to avoid obstruction of the light. The technical details of the TRL pilot scale lighting installation are summarised in Table 4.1. TRL Limited 62 PPR043

73 Table 4.1 Details of TRL pilot scale lighting installation HPS CMH No of columns 5 Spacing Lamp height 30m 9m Luminaire Thorn Decostreet 2 Lamp type Ballast GE Lucalox (150W, TT envelope, E40 cap, CCT ~ 2000 K, Ra ~ 27) Luxmate Dimtronic Prototype GE (150W, TT envelope, E40 cap, CCT~4200 K, Ra ~ 88) Knobel DALI step-dim 150/70 ballasts Powerline transceiver DALI controller Luxmate OLC Combox Transceiver (CTS) capable of controlling up to 127 luminaire controllers Luxmate OLC DALI Lantern Dim Controller Lamp types CMH and HPS The HPS and CMH lamps used in the trial were as follows. The HPS lamp was a standard GE Lucalox lamp from the GE lighting catalogue, with a specified (100 hours) output of 15,000 lumens at 150W (= 100 lumens per watt). The enclosure is a standard tabular TT in clear glass with an E40 mounting cap (Figure 4.4). Published data show that the output decreases to 80% of initial (100 hour) at between 13,000 and 21,000 hours depending on operating conditions, with 80% of lamps expected to survive to between 15,000 and 21,000 hours. The five lamps for the trial were aged to 100 hours prior to installation. This is a standard age for the assessment of initial output. The CMH lamp was a prototype. Initial measurements by GE indicated an initial (100 hours) average output of 13,967 lumens at an average of 143W (~ 98 lumens per watt). The enclosure was a standard tabular TT in clear glass with an E40 mounting cap (Figure 4.5). Results from GE show that at 500 hours (the next age point recorded), for similar lamps burning horizontally and operated continuously at 150W for 50%, and 100W for 50% of the time, average output reduced by about 6%. After an initial period of system testing following installation, the lamps were operated for no more than a further 60 hours during the trial. The output was therefore not expected to alter by more than about 1% over the duration of the trial. This was confirmed by means of illuminance measurements. TRL Limited 63 PPR043

74 Figure 4.4 HPS 150W lamp as used in the TRL trials Figure 4.5 CMH 150W lamp as used in the TRL trials 4.2 Selection of observers It was decided to select observers of both sexes and from three age ranges, with the aim of giving a wide sample of the range of physical and visual acuity likely to be encountered in the driving public. Participants were selected from a database of external volunteers held by TRL for the purpose of driver behaviour studies. In order to check the visual acuity of observers a basic eye test was carried out to confirm that their eyesight was of an acceptable standard. The observer was asked to read a standard Snellen letter chart, as used by opticians, using both eyes, from a distance of 6 metres. Spectacle wearing observers wore the spectacles they normally used for driving during this test. All observers selected for the trials were found to have at least average visual acuity. Average visual acuity corresponds to line 6 of the Snellen chart. The letters on this line of the chart are 8.75mm high and have a stroke width of 1.75mm, which corresponds to an angle of resolution of 1 of arc when viewed from a distance of 6 metres. Those with scores of 5 were able to read further down the chart, and hence were deemed to have better than average visual acuity. Table 4.2 presents the results of the eye test, age and sex of each observer. TRL Limited 64 PPR043

75 Table 4.2 Observer sex, age and Snellen scores Observer No Gender Snellen Score (6/n) Age (years) 1 Female Female Male Male Male Female Central vision experiment The central vision experiment was designed to demand the use of the central visual field. Although, it is accepted that the whole visual field was being used, including the central field, finding a small target on the carriageway is considered to be primarily a central visual task. The task involved observers driving at a set speed along a length of carriageway and indicating when they could detect a small target placed on the road surface. The reasons behind the selection of vehicle speed, target and location are discussed in the following sections. Further information on specific experimental details are provided in Section Selection of vehicle speed The selection of a vehicle speed for the experiment was limited by practical considerations which included those imposed by the trial road layout, the accuracy achievable (in terms of calculating vehicle position for a particular speed) and the amount of time available for the observer to make the observation. The approach to the trial stretch was fairly short (see Figure 4.2). Initial trials found that acceleration to 30km/h, with speed being relayed by a TRL researcher, typically allowed enough time for the observer to concentrate on the visual task before a target became visible. This speed, equivalent to approximately 8.3 metres per second gave acceptable accuracy in terms of vehicle location. The vehicle used in the trials was a new Ford Focus estate car with standard halogen headlamps Selection of target and location Participants were asked to indicate when they observed a small target, having a measured diffuse reflectance of 25%, placed on the road surface. In previous street-lighting research, including experiments for TRL reported by Cooper and Diver (2000), a 200mm diameter flat target has been used. However work by Lecocq (1998), and subsequently by Bacelar and Beaucamp (2002), used a spherical target considered to better represent real objects, which show graduations of luminance. In order to facilitate a decision on the use of flat or spherical targets a number of preliminary trials were carried out to identify the most appropriate. The result of these trials was the selection of the 200mm spherical target. This is described further in Section The number and location of target positions on the road surface was limited both by the necessity to keep the running lane free of obstructions, and the time needed to carry out a complete set of observations. The target area was the length of road between Columns 3 and 4. This permitted the observers the maximum possible distance and time to adapt to the luminance level of the lit area, while maintaining a fully lit area as background to each target. TRL Limited 65 PPR043

76 Six target positions were chosen. The positions were distributed laterally near the kerb, the road centreline and the centre of the opposite lane on three longitudinally spaced lateral lines, as shown in Figure 4.6. To maintain the same relative target position for each lamp, two positions were defined for each target, one for observations under the CMH lamps and one for the HPS lamps. These accommodated the longitudinal offsets of ±0.45m of the HPS and CMH luminaires from each column. It should be noted that all target observation data (presented in Section 4.8.1) have been adjusted to take account of this. To facilitate correct placement of the target, all the positions were marked on the road surface by steel road nails inserted flush with the surface. Figure 4.6 Target locations for central vision experiment TRL Limited 66 PPR043

77 4.4 Peripheral vision observations The aim of this part of the trial was to assess observers peripheral visual reaction times under each of the different lighting conditions. Experimental details are given in Section In this task, observers indicated when they detected a change in target luminance under various lighting conditions. This change in luminance was achieved using a rectangular target changing from black to grey (25% reflectance). The equipment used for these observations is shown in Figure 4.7. It consisted of the flip dot target shown on the left and the fixation target shown on the right. It should be noted that the fixation target is not shown in its actual operating position, which was remote. Further details of this experiment are provided in Section Figure 4.7 Flip-dot panel (left) and fixation target (right, in operation placed at the far end of the trial road) 4.5 Trial parameters The parameters and levels detailed in Table 4.3 were identified for the central vision and peripheral vision experiments. This resulted in a total of 864 observations for the central vision experiment. Each observer attended one session under each of the four lighting conditions, during which six observations were made of each of the six target positions. This required 24 sessions. The peripheral vision observations were undertaken mid way through each session. The number of levels and repeats of each variable were limited by practical considerations. Trials could only be carried out when the road surface was dry, in order to keep the luminance properties of the road constant and comparable with the design r-tables, which are calculated for dry roads only. The lighting configuration of the vehicle was also kept constant, with headlamps on and dipped. The contribution of vehicle headlamps to target detection was investigated and is discussed in Section TRL Limited 67 PPR043

78 Table 4.3 Parameters for track trial experiments Property Central vision experiment Peripheral vision experiment Lamp types Power CMH HPS 150W 95W Surface type/condition surface dressing dry surface dressing dry Headlamp setting dipped dipped Opposing headlamps none none No of subjects Male Female Age range Target positions 6 2 Repeats of each observation 6 10 Total number of observations TRL Limited 68 PPR043

79 4.6 Measurements of SRS trial lighting system components In addition to the specific target related measurements (described later in Sections 4.8 and 4.9), measurements of the lighting system are described in the following sections. These consisted of luminance and illuminance measurements used to assess road surface characteristics, and those used to assist in the selection of target type. In addition, power measurements were undertaken to assess electricity consumption for both CHM and HPS, both dimmed and undimmed. Luminance measurements were taken using a photometric camera, positioned so that various images of the road surface could be achieved. Technical details relating to this camera are presented in Appendix G.1. Figure 4.30 shows examples of luminance photographs for the two lighting types at 150W. The different colours represent various levels of luminance (in terms of candelas per square metre) as shown by the scales on the left of the images. Illuminance measurements were collected using a Jeti Specbos 1200 spectroradiometer, which is a compact instrument capable of 5nm wavelength resolution over the range 380nm to 779nm. The software provided calculates average radiance and a range of photopic parameters, including luminance (illuminance with a cosine attachment), colour rendering indices, correlated colour temperature and CIE colour coordinates. Full technical details are provided in Appendix G.2 Power measurements were undertaken using a Metrahit 29s digital multimeter/powermeter with clipon current transformer. Further details are provided in Section Target area measurements Luminance measurements Examples of the luminance images of the trial road in the target area (between Columns 3 and 4) are presented in Figure These luminance photographs were used to determine the overall uniformity and longitudinal uniformity for the road surface between Columns 2 and 4, and 3 and 4. The pictures were taken from a position on the centreline of the left-hand lane, at a height of 1.3 metres and a distance, of 60 metres to the centre of the former area and 75 metres to the centre of the latter area. This resulted in angles of view (below the horizontal) of 1.25 for the area having its centre at Column 3, and 1 for the area having its centre mid-way between Columns 3 and 4. These are within the range of 1 ± 0.5 specified in BS EN 13201: Part 4. For reference, longitudinal and overall uniformity are defined in BS EN13201: Part 2: 2003 as follows: Longitudinal uniformity (of road surface luminance of a driving lane) ratio of the lowest to the highest road surface luminance found in a line in the centre along a driving lane Longitudinal uniformity (of road surface luminance of a carriageway) (U l ) lowest of the longitudinal uniformities of the driving lanes of the carriageway Overall uniformity (of road surface luminance) (U o ): ratio of the lowest to the average value Measurements were made on the images using the LMK2000 software. To obtain the overall uniformity, measurements were obtained within a polygonal region on each luminance photograph that covered the required area as closely as possible. The longitudinal uniformity was calculated along a line drawn along the centre of the left hand lane, to match the driven path of the car. These procedures were not precisely according to the standard which also requires a measurement taken TRL Limited 69 PPR043

80 from a position in the centre of the right-hand lane and the lowest of the results to be taken. The longitudinal uniformity obtained was therefore that of road surface luminance of a driving lane, and not U l, as defined above. Examination, on the same images, (from the position in the left hand lane) of the approximate longitudinal uniformity in the right hand lane suggests that this is unlikely to have been worse than that in the left hand lane, so the error in the measurement of U l is likely to be small. Table 4.4 shows the resulting coefficients. The specification for UK roads splits roads into various categories and specifies acceptable minimum values of uniformity for each category. Table 4.4 Uniformity data for CMH and HPS lighting Average luminance Overall uniformity (both lanes from left-hand lane) Longitudinal uniformity in the left-hand lane Lamp and Power CMH 95W CMH 150W HPS 95W HPS 150W Region Column 2 to Column 4 Column 3 to Column 4 Column 2 to Column 4 Column 3 to Column 4 Column 2 to Column 4 Column 3 to Column 4 Column 2 to Column 4 Column 3 to Column 4 L (cd/m 2 ) U Table 4.5 is reproduced, in reduced form, from BS EN and shows the uniformity required for the ME series of lighting classes. These are intended for drivers of motorised vehicles on traffic routes of medium to high driving speeds. The values obtained for the overall area in which the target, for the central vision experiment, was placed are acceptable in terms of overall uniformity, for all UK road types (Table 4.5). However, when the measured longitudinal uniformity is taken into account the lighting is only acceptable for roads of type ME4b or lower. This would restrict the suitability to comparatively lightly trafficked roads. TRL Limited 70 PPR043

81 Table 4.5 Extract from Table 1A, BS EN ME-series of lighting classes Class Luminance of the road surface of the carriageway for the dry road surface condition L in cd/m 2 (minimum maintained) Overall uniformity (Uo ) (minimum) Longitudinal uniformity ( U l ) (minimum) ME ME ME3a ME3b ME3c ME4a ME4b ME ME Illuminance measurements Horizontal illuminance measurements were taken at the intersections of a grid at 10m intervals longitudinally and 2.25m laterally on the road surface in the target area. The spectroradiometer was mounted, with its optical axis vertical, just above the road surface on a miniature tripod and fitted with an illuminance attachment. The data represent the light from the lamps falling on the road (as opposed to light reflected from the road) in units of lux (lumens per square metre). These are presented in Table 4.6. Figure 4.8 and Figure 4.9 show the measured illuminance distribution in graphical form for each of the lighting schemes. TRL Limited 71 PPR043

82 Table 4.6 Illuminance data for CMH and HPS lamps at 150 and 95W Lamp type and power Lateral position across road (m) Illuminance (lux) Distance from Column 3 towards Column 4 (m) CMH 150W CMH 95W HPS 150W HPS 95W TRL Limited 72 PPR043

83 CMH 150W to to to to to to to 15 9 to 12 6 to 9 3 to 6 0 to Illuminance (lx) Distance from column 3 (m) Illuminance (lx) offset from kerb (m) CMH 95W to to to to to to to 15 9 to 12 6 to 9 3 to 6 0 to Illuminance (lx) Distance from column 3 (m) Illuminance (lx) offset from kerb (m) Figure 4.8 Illuminance data for CMH lamps at 150W and 95W HPS 150W to to to to to to to 15 9 to 12 6 to 9 3 to 6 0 to Illuminance (lx) Distance from column 3 (m) Illuminance (lx) offset from kerb (m) HPS 95W to to to to to to to 15 9 to 12 6 to 9 3 to 6 0 to Illuminance (lx) Distance from column 3 (m) Illuminance (lx) offset from kerb (m) Figure 4.9 Illuminance data for HPS lamps at 150W and 95W TRL Limited 73 PPR043

84 4.6.2 Target design and selection for central vision experiment Target luminance and illuminance It was intended that the targets used for the central vision trial should have a reflectance level of around 20%. The targets were painted to achieve a colour match as close as possible to a photographer s grey reference card, with a quoted reflectance of 18%. This reference is used to set exposure in photography, and was chosen as a standard for this experiment following a review of previous research on target visibility (Bacelar and Beaucamp, 2002; Lecocq, 1998). Luminance and illuminance data were collected in the laboratory for the spherical target and are presented in Table 4.7. The actual diffuse reflectance measured on the targets was between 26.0% and 26.4%. Table 4.7 Laboratory reflectance measurements for spherical target used in the trial Lamp and power Illuminance (lux) Luminance (cd/m 2 ) Diffuse reflectance (%) CMH (150W) HPS (150W) Note: Matt paint was used to achieve as diffuse a response from the target as possible Comparison of flat and spherical targets It was noted in reports by Lecocq (1998) and Bacelar and Beaucamp (2002) that both plane circular and spherical targets had been used previously in other similar observation experiments. Lecocq indicated that spherical targets were more representative of real objects. For the central vision experiment it was necessary to evaluate which type of target was most appropriate, either a circular (plane) or a spherical target. Therefore preliminary trials, using TRL staff, were undertaken with both a 200mm diameter circular (plane) target and a 200mm diameter spherical target. In general, it was found that the plane circular target was very difficult to see compared to the spherical target with, in some cases, detection distances of less than 15m compared to over 100m using the spherical target. Under some conditions it was observed that the plane circular target was not detected until illuminated by the car s headlamps. The spherical target could be seen outside the range of the (dipped) headlamps. This target was therefore deemed to be more suitable for the experiment, which aims to test visual performance under the lamps. Luminance photos of each target were also taken from 50 metres at a height of 0.9 metres (to give the required angle of 1 ) in order to compare the luminance of each target with that of its background (see Figure 4.10). This identified that, for the plane circular target, no significant differentiation of target and background was apparent and confirmed that this target type was unsuitable. On the basis of these investigations, it was decided to adopt the spherical target for the central vision experiments. TRL Limited 74 PPR043

85 Figure 4.10 Comparison of flat and spherical targets in position C (LHS) and D (RHS) under HPS 150W Effect of headlamps During the central vision and peripheral vision trials the car was used with its headlamps dipped to create as realistic an environment for the observers as possible. To assess the amount of light falling on the target from the vehicle s dipped headlamps and its potential to affect the ability of the observers to detect targets, both the spherical and flat circular targets were positioned 50m from the car to simulate positions A and E, and C and D (see positions in Figure 4.6). Luminance photographs were then taken at a distance of 50 metres from the targets with the street lighting switched off. It was assumed that, for headlamp illumination, positions B and F (along the centre line of the trial road) were equivalent to target position C. Table 4.8 shows the mean, maximum and minimum luminance values obtained from the luminance photographs for all six target positions. TRL Limited 75 PPR043

86 Table 4.8 Luminance measurements under vehicle headlamps at 50m Target Type Target position Mean (cd/m 2 ) Luminance of whole target Maximum (cd/m 2 ) Minimum (cd/m 2 ) A Flat B,C,F D E A Spherical B, C, F D E Analysis of the data reveals that the flat target reflected more light from the headlamps back to the camera, and hence the observer, than the spherical target. This is to be expected as the geometry of the spherical target will disperse light. From preliminary trials it also seemed probable that at the distances at which the observers detected the targets, little of the light attributable to the headlamps was being reflected back from the targets. The luminance levels from the headlamps at target positions B, C, D, and F were very low (around 0.04 cd/m 2 ) and therefore unlikely to have had any measurable effect on the ability of the observers to detect the target. As would be expected the ratio of maximum to minimum luminance was greater for the spherical target than the flat circular target for all positions except D. Target position D was the position least affected by light from the vehicle headlamps giving very low luminance levels. Dipped headlamps illuminated the targets at positions A and E, situated to the nearside of the car, significantly more than the remaining four target positions on the offside of the vehicle. In general, it was found that targets placed at positions A and E were the most difficult to detect under all four lighting conditions employed in the study. This would suggest that the effect of light from the headlamps falling on the targets at positions A and E did not significantly improve the ability of observers to detect targets placed in these positions. Dipped headlamps were also used during the peripheral target detection tests. In view of the height of the flip dot target and its angle to the observer, and hence the headlamps of the car, it was considered that any effect due to the headlamps would (a) be constant and (b) be small in comparison to the output from the lamps Peripheral vision experiment Details of the equipment and procedures used to determine the observer s ability to detect the sudden change in colour of a target panel from black to grey placed first at 15 o ( kerb ), and then at 25 o ( back of path ), to the nearside of the stationary car are given in Section The target used for this experiment consisted of a flip dot panel that could be changed remotely from black to grey. The grey face of each flip dot was painted with the same paint as the spherical target and hence reflected about 25% of the light falling on it. Luminance photographs of the target under HPS lighting at full power, in both grey and black positions, and at both 15 and 25 angles are shown in Figure 4.11 and Figure 4.12 (note that false-colour luminance scales vary). TRL Limited 76 PPR043

87 The luminance scales used in the photographs were automatically applied by the analysis software. It should be noted however, that the peak values for the 15 o position are approximately 2.6 times those for the 25 o position. Figure 4.11 Luminance photos of flip dot panel at 15 viewing angle (black in top image, grey in bottom image) for HPS 150W TRL Limited 77 PPR043

88 Figure 4.12 Luminance photos of flip dot panel at 25 viewing angle (black in top image, grey in bottom image) for HPS 150W TRL Limited 78 PPR043

89 Table 4.9 Luminance measurements of flip dot panel Luminance Lamp type Power and flip dot colour 15 viewing angle ( kerb ) (cd/m 2 ) 25 viewing angle ( back of path ) Ratio (15 to 25 viewing angle) 150W Grey W Black HPS 150W Ratio (Grey to Black) W Grey W Black W Ratio (Grey to Black) W Grey W Black CMH 150W Ratio (Grey to Black) W Grey W Black W Ratio (Grey to Black) Table 4.9 shows the luminance values obtained for the flip dot area of the target panel at both peripheral angles (15 o and 25 o ), and the four types of lighting used in the trials. The measurements showed that the ratio of luminance between the flip dots in their grey and black positions was between 1.6 and 3.0. Because the target panel was moved to the back of the footpath to create the 25 o angle, the luminance level of the target was significantly reduced by a factor of between 1.7 and 2.8. This may have affected the observer s detection times for the 25 o angle position. However, it was felt that moving the target panel created a more realistic scenario for the observer than moving the car to the centre of the road. In order to assess the contrast between the flip dot target area and its surrounding background luminance, measurements of several areas are shown in Figure 4.13 to Figure The areas measured were rectangular in shape and covered the flip dot area, the black masked area, and the background both above and below, and to the right and left of the target for three lamp conditions. TRL Limited 79 PPR043

90 fd 5 3 luminance cd/m Flip 1 dot region number black kerb grey at kerb grey at back of path black at back of path Figure 4.13 Luminance under HPS at 100% light output fd luminance cd/m Flip 1 dot region number black kerb grey at kerb grey at back of path black at back of path Figure 4.14 Luminance under HPS at 50% light output TRL Limited 80 PPR043

91 fd 5 3 luminance cd/m Flip 1 dot region number black kerb grey at kerb grey at back of path black at back of path Figure 4.15 Luminance under CMH at 100% light output 0.3 luminance cd/m fd Flip 1 dot region number black kerb grey at kerb grey at back of path black at back of path Figure 4.16 Luminance under CMH at 50% light output The measurements for the HPS lamps at full power indicated that when the flip dots were in the grey position their average luminance was higher than that of any of the surrounding areas (Figure 4.13). For the flip dots under both CMH 100% and CMH 50% lamps this was found not to be the case (Figure 4.15 and Figure 4.16). The background below the target had a higher average luminance than the grey flip dots. This may have affected the observer s detection times under the CMH lamps. A possible explanation for the brightness of the background below the target could be that the whiter light from the CMH lamps, containing significantly more shorter wavelength light (blue end of the visible spectrum), was reflected from the green vegetation making up the background below the target, more than that from the (closer to monochromatic) HPS light. Black masking was used to cover the left side of the target panel. This modification was necessary as the electronic timer used to measure the observers detection times was triggered by the centre column of dots as they flipped from black to grey. Left unmasked, the observer would have been exposed to part of the black to grey target colour change for a few milliseconds before the timer was triggered, leading to unrealistically short detection times. TRL Limited 81 PPR043

92 4.6.4 Power consumption Power consumption of the whole system was measured at the control box, using a digital multimeter/powermeter with dedicated current clamp, enabling current measurements without the requirement to break the circuit. The meter simultaneously measured voltage via pre-installed probe connections in the control box. The power was displayed on the screen of the meter, in watts. This measurement enabled confirmation of the system s power requirement, and was used to check the correct function of the ballast by indicating that the expected power was being drawn. A measurement of the reactive power demand was also taken. The results are shown in Table I1 of Appendix I. With the system in standby the measurements indicated a comparatively large reactive power requirement with no output from the lamps, ie with only the ballasts in circuit. Otherwise, the power requirements were within the expected range. 4.7 Experimental procedure The experiment involved two major tasks. As described in Section 4.3, the first aimed to determine the observers ability to detect a target, placed on the road surface, whilst driving a vehicle at constant speed. The second task was performed with the vehicle stationary in the centre of the left hand lane, and measured the observers reaction times in response to a change of target luminance in their peripheral fields of vision. In addition to these tasks, the observers were also asked to complete a questionnaire designed to elicit their subjective response to each lighting condition Central vision task The aim of the central vision task was to ascertain whether there were significant differences in the distances at which targets could be detected under each of the four lighting conditions. After preliminary trials it was decided to perform the driving task at 30 km/h. Therefore, a system was required that could accurately locate the vehicle on the trial road when a target was observed. To work out the location of the vehicle, and thus the distance from the target, time and distance data were recorded relative to a fixed marker post at the start of the run. The data acquisition equipment used consisted of a custom assembly incorporating an electronic (0.01 second) timer, a steering wheel mounted switch, an optical sensor switch and a Correvit optical speed sensor interfaced to a laptop computer. Both sensors were mounted together on the nearside front door of the trial vehicle (see Figure 4.17). The Correvit sensor faced the road surface vertically and the optical switch was positioned so as to trigger at a retro-reflective marker post placed at the start of the trial run. This marker post was positioned at point B (see Figure 4.6). The optical switch started both the Correvit and the timer. On detecting the target, the observer pressed the steering wheel switch to stop the timer. The measured elapsed time was used in conjunction with the Correvit data to calculate the distance travelled beyond the trigger post. The Correvit system was set to record time, distance, velocity and acceleration data at intervals of 1m. At the set speed of 8.3m/s the timing allowed a distance resolution of close to 0.1 metres, by linear interpolation between the two nearest Correvit records. A TRL researcher in the rear seat of the car operated the data collection system, recorded the reaction times and target positions and saved data files for each individual run. A second researcher placed the target, in a random sequence, in one of the six marked positions and relayed, via radio, its position after each run had been completed. Figure 4.18 shows a block diagram of the system. In some instances no target runs were carried out. It was noted that if observers did press the timer stop switch in this situation they indicated their mistake. This gives some confidence that false detections were not included in the data. Runs in which the observer made a mistake were repeated. Observations of each target position were repeated six times, in an unpredictable order, during each session. TRL Limited 82 PPR043

93 Figure 4.17 Marker post, Correvit speed sensor and optical switch fitted to vehicle Figure 4.18 Block diagram of central target data collection system TRL Limited 83 PPR043

94 Experimental details Each trial started at least one hour after sunset. Observers were briefed on the purpose of the trial and the tasks they were to carry out. In addition the observers had been informed of the need to avoid alcohol or other drugs which might affect their performance during the trial. The observers were given the opportunity to familiarise themselves with the vehicle and carry out a number of trial observations. These observations were not recorded and therefore did not form part of the data set. The observer drove past point A, turning onto the trial section (see Figure 4.2). As the observer accelerated to 30 km/h their speed was verbally relayed by the TRL researcher. This allowed the observer to concentrate on target detection. Due to the limited number of observers it was not possible to completely design out any learning effects. It was decided to present the lighting conditions in the same order for all observers. Each observer started with the full-power CMH. The sequence of the set of trials that was planned for each observer is presented in Table Figure 4.19 and Figure 4.20 show the vehicle and spherical target under the CMH and HPS lighting. Table 4.10 Session Trial sequence Lamp and Power Setting (W) 1 CMH 150W 2 HPS 150W 3 CMH 95W 4 HPS 95W Figure 4.19 Trial vehicle and spherical target under CMH lighting TRL Limited 84 PPR043

95 Figure 4.20 Trial vehicle and spherical target under HPS lighting Peripheral vision experiment The peripheral target used was a panel of flip-dots, similar to those used to display destinations on the front of buses. This was a rectangular matrix of flip-dots (20 wide x 14 high) measuring 300mm wide by 210mm high. The flip-dots are regular octagons, providing a ratio of flip area to total area of about One side of each flip-dot was matt black in colour, the other side was painted with the same paint used for the spherical target. On triggering, the dots flip in a sequence from left to right. This interval was measured using a digital video camera. The complete change took 5 frames at 15 frames/sec, giving an interval between 0.2 and 0.4 seconds. The trigger point for timing was at the centre of the panel. To reduce the delay between initiation and completion of flipping, and to bring the timing trigger point to the left edge of the effective area, the left half of the panel was masked by black card. This reduced the time for the complete flip by a factor of two. The flip could be initiated by the TRL researcher in the back seat of the car, by means of a remote control box. The remote control box was designed by TRL and included an electronic timer accurate to one hundredth of a second.. The timer was stopped by a button held by the observer which was pressed on detection of the change in the target. The arrangement of the target and timing equipment is shown schematically in Figure 4.21 The flip-dot panel was mounted with the bottom edge at a height of one metre above the road on a tripod placed midway between (15 metres from) Columns 3 and 4. The car was positioned in the centre of the lane so that the drivers eyes were 10m longitudinally from the flip-dot position (Figure 4.22). In Position 1, the target was placed 2.7 metres (15 ) laterally from the observers sight axis. In Position 2, the target was placed 4.7 metres (25 ), laterally from the observers sight axis. Position 1 was centred over the kerb at the edge of the road and Position 2 was 2 metres further from the kerb. To ensure that the observer was looking in the correct direction an illuminated fixation target was placed at the end of the trial road (the fixation target is shown in Figure 4.7). The observer was asked to continuously relay the randomly changing colour of the target to the TRL researcher. This ensured that the observer was continually fixated on the distant target and the flip-dot panel was in the peripheral region of vision. TRL Limited 85 PPR043

96 Following a change in flip dot luminance, the observers reaction time was recorded. Ten repetitions were carried out with the target in each position in order to increase overall accuracy. The car s headlamps were on and dipped during this procedure. Figure 4.21 Block Diagram of Peripheral Target Data Collection System TRL Limited 86 PPR043

97 Figure 4.22 Schematic layout of flipdot panel in relation to carriageway Questionnaire A questionnaire was given to observers on completion of each trial. This allowed the participants to give feedback on each lighting condition, in terms of lighting level and uniformity, amount of glare, and overall impression of the lighting. The questionnaire comprised a set of three tick-box questions with nine-point scales, and an area for general comments (Appendix H). The observers responses are summarised in Section Supplementary measurements During each trial additional measurements were taken for the purpose of confirming the consistency of the lighting conditions. These are detailed in the following sections. TRL Limited 87 PPR043

98 Illuminance The horizontal illuminance vertically beneath the CMH or HPS lamp in use (Column 1) was measured during every session to confirm the consistency of light output during the trial. Similar measurements were made under other lamps during a limited number of sessions, to check relative lamp output Luminance For each trial, luminance photographs were taken from a position adjacent to Column 1, in the middle of the left hand lane at a height of 1.3m. This gave the standard 1 viewing angle at the centre of the target area between Columns 3 and 4, which was 75m away from the camera position. This enabled a comparison, between trials, of the luminance of the road surface. Four white LED marker lamps were used at the kerb, placed on both sides of the road adjacent to Columns 3 and 4. The marker lamps provided a means of defining the four corners of a grid of known dimensions on the road surface, thus allowing potential for further analysis of the luminance images. 4.8 Results Central vision The central vision experiment collected target detection distance data for each of the observers and each of the lighting conditions. This resulted in a range of distances being recorded which have been analysed in terms of observer, lamp type, power setting, age, sex and visual acuity. Due to the natural variability of the data, caused by human and other factors, simple analysis in terms of mean detection distances can be misleading. This is because one or two outlying results can adversely alter the mean. Results of the trials are therefore presented as boxplots. This type of plot gives an indication of the spread, interquartile range and median values of the presented data. These attributes are illustrated in Figure The boxplot comprises two whiskers, a box and a median line. The majority of the data resides within the range of the whiskers (significant outliers are excluded but in general can be viewed as maximum and minimum values). The box contains the middle 50% of the data (interquartile range). The benefit of this type of plot is that it gives a good indication of the range and significance of the plotted data and allows a rapid visual comparison of complicated data sets. Figure 4.23 Boxplot attributes General overview In order to find any trends in visual performance under the different lighting conditions, Figure 4.24 presents data for all observers, target positions and lamp type and power combinations (it should be noted that in this and following figures a greater target detection distance represents better visual performance). From this it is evident that, for the majority of target locations, the interquartile ranges of detection distance for individual lamp and power combinations show a high degree of overlap and the median values are broadly similar. The notable exceptions are target position E and to a lesser extent target position A, where the interquartile ranges and median values diverge. TRL Limited 88 PPR043

99 Distance from target when observed (m) Lamp and Power CMH 150W CMH 95W HPS 150W 0 A B C D E F HPS 95W Target ID Figure 4.24 Boxplot showing variation in detection distance for each lamp and power setting Presented in this way, the data suggest that for most of the targets there was no significant difference in target detection distance between CMH and HPS at either power level or between full power and dimmed for either lamp type. However, the target detection distances for positions A and E under both levels of CMH, show that these were more difficult to detect. Possible explanations are discussed further in Section The next stage in the analysis was to investigate the influence of observer, age, sex and visual acuity and, if possible, remove the variance due to these factors from the data, which might reveal greater differences in visual performance between the four lighting conditions Assessment of observer effects An analysis of the effect of the observer for all lamp and power combinations is shown in Figure It is apparent that the data fall into two distinct categories. The first contains observers 1, 2 and 3 whilst the second consists of observers 4, 5 and 6. This suggests that some of the variance in the data is observer dependent. This agrees with the expectation that both reaction times and visual acuity deteriorate with age (Department for Transport, 2000). Further analysis of the data, categorising the observers into their respective age bands reveals a downward trend in detection distance of all targets with age of observer (Figure 4.26). This trend represents poorer detection distances for observers of a greater age. This effect is likely to be a combination of both changes in visual acuity and generally increased reaction times. This downward trend is also apparent when a measure of visual acuity is used to discriminate the data. Figure 4.27 presents, for each target position, the detection distances of all the observers with respect to their Snellen scores. As discussed earlier poorer vision is indicated by a higher Snellen score. Figure 4.27 shows differences in the median detection distance, for Snellen scores of 5 and 6, in excess of 20m to 25m. At the trial speed of 8.3m/s this equates to between 2.4 and 3.0 seconds difference in median detection time. As both visual acuity and physical reaction time reduce with age this is to be expected in an experiment of this type. That this effect can be discriminated from the data demonstrates that, despite the considerable scatter in the data, the analysis is capable of showing any significant effects present. There were no discernable differences in detection distance between the observers of different sex (Figure 4.28). TRL Limited 89 PPR043

100 Distance from target when observed (m) Observer A B C D E F 6 Target ID Figure 4.25 Variation in detection distance for each observer Distance from target when observed (m) Age Band 20 to to 50 0 A B C D E F 60 to 70 Target ID Figure 4.26 Variation in detection distance with age TRL Limited 90 PPR043

101 Distance from target when observed (m) Snellen score 20 6/5 0 A B C D E F 6/6 Target ID Figure 4.27 Variation in detection distance with Snellen score Figure 4.28 Variation in detection distance with sex of observer Removal of observer dependent factors In order to remove the effects of differences in age and visual acuity from the data a statistical method called UNIvariate ANalysis Of Variance (UNIANOVA) was used. This allowed the effects of nonobserver dependent factors, such as lamp and power to be seen more clearly. TRL Limited 91 PPR043

102 The analysis used a hierarchic analysis of variance model, which fitted subject at the top level. In this way the variation between subjects was controlled, resulting in a more precise estimate of the residual error term, i.e. the 'noise' due to between-subject variation was eliminated. It is reasonable to assume that subjects responded in a consistent way on each trial and so the differences between lighting types/levels could be assessed with greater precision once the extra noise was eliminated. The differences between mean values are not affected, just the residual error. It was known that subjects had slightly different visual acuities, so acuity was fitted as a covariate in one model. It was statistically significant but only made a minor difference to the average distance measures; the overall results were not significantly changed. It is accepted that more subjects would have been desirable, but the robustness of the results suggest that the conclusions are defensible. The analysis provided an indication of the remaining differences in target detection distance between the lighting conditions and whether they are significant at the 95% level (also known as 95% confidence interval). The results of analysis using this method (using SPSS software) are presented in Table 4.11 and Figure On the basis that differences are only valid if the 95% confidence intervals do not overlap for particular combinations, it is evident from Table 4.11 and Figure 4.29 that significant differences in mean detection difference between the lighting conditions exist for Targets A, B, C and E. The extent of these differences are presented in Table 4.12 (measured relative to CMH 150W). It is evident that small differences, less than 6m, occur for target positions B and C and greater differences occur for positions A and E. However, the significance of the differences at positions B and C is dependent on the lamp and power combination against which an assessment is made. For instance if differences are measured relative to CMH 95W none of the differences are significant, as the 95% confidence intervals overlap. It is only in target positions A and E that significant differences (at the 95% confidence level) are apparent between CMH and HPS lighting. It is worth noting that these positions were located on the edge of the footpath and on the carriageway adjacent to the kerb, respectively. The backgrounds to these target positions were less uniform. Further analysis was therefore required to ascertain whether the differences apparent at positions A and E were related solely to lamp type or whether other factors may have influenced the results. This is described in the following section. TRL Limited 92 PPR043

103 Table 4.11 Results of univariate analysis Target position Lamp type Power (W) Significant (relative to CMH 150W)? Mean observation distance (m) 95% confidence interval of the mean Upper bound (m) Lower bound (m) CMH A CMH 95 N HPS 150 Y HPS 95 Y CMH B CMH 95 N HPS 150 Y HPS 95 Y CMH C CMH 95 N HPS 150 Y HPS 95 Y CMH D CMH 95 N HPS 150 N HPS 95 N CMH E CMH 95 N HPS 150 Y HPS 95 Y CMH F CMH 95 N HPS 150 N HPS 95 N Note: A higher observation distance indicates the target was observed earlier TRL Limited 93 PPR043

104 Distance from target when observed (m) Upper Bound Lower Bound Mean 0 A A A A B B B B C C C C D D D D E E E E F F F F CMH CMH HPS HPS CMH CMH HPS HPS CMH CMH HPS HPS CMH CMH HPS HPS CMH CMH HPS HPS CMH CMH HPS HPS 150W 95W 150W 95W 150W 95W 150W 95W 150W 95W 150W 95W 150W 95W 150W 95W 150W 95W 150W 95W 150W 95W 150W 95W Target and Lamp and Power Figure 4.29 Result of univariate analysis of variance with observer effects removed On the basis that differences are only valid if the 95% confidence intervals do not overlap for particular combinations, it is evident from Table 4.11 and Figure 4.29 that significant differences in mean detection difference between the lighting conditions exist for Targets A, B, C and E. The extent of these differences are presented in Table 4.12 (measured relative to CMH 150W). It is evident that small differences, less than 6m, occur for target positions B and C and greater differences occur for positions A and E. However, the significance of the differences at positions B and C is dependent on the lamp and power combination against which an assessment is made. For instance if differences are measured relative to CMH 95W none of the differences are significant, as the 95% confidence intervals overlap. It is only in target positions A and E that significant differences (at the 95% confidence level) are apparent between CMH and HPS lighting. It is worth noting that these positions were located on the edge of the footpath and on the carriageway adjacent to the kerb, respectively. The backgrounds to these target positions were less uniform. Further analysis was therefore required to ascertain whether the differences apparent at positions A and E were related solely to lamp type or whether other factors may have influenced the results. This is described in the following section Potential reasons for differences in detection distances at target positions A and E In broad terms the factors likely to affect the detection distances at target position A and E can be categorised as follows: illumination of the road surface and target, observer effects such as age and visual acuity, contrast differences under each of the lighting conditions. TRL Limited 94 PPR043

105 Table 4.12 Extent of non overlap of 95% confidence intervals Target Lamp Power (W) Extent of non overlap for 95% confidence interval (a) (b) (m) A B C E HPS HPS HPS HPS Notes: (a) For consistency, differences are expressed relative to CMH 150W. (b) A positive value indicates the target was seen earlier, i.e. further from the target Illumination of road surface and target In order for comparisons and hence differences to be valid, the overall distribution of light falling on the trial road under CMH and HPS lighting would need to be broadly comparable. Figure 4.30 presents two luminance photos of lighting under CMH and HPS lamps, both at a power setting of 150W. It can be seen that the regions with a (photopic) luminance of 2cd/m 2 cover different areas of the road surface. For the CMH lamps this area can be seen to extend from the middle of the right hand lane to approximately the middle of the left hand lane, whereas, for the HPS lamps this area extends nearly the full width of the carriageway and onto the footpath at the side of the road. This is a possible explanation why only targets placed on the left hand side of the trial showed any significant difference in detection distance between CMH and HPS. It should be noted that the sizes of the different areas are to some extent due to the different photopic luminous outputs of the two types of lamp. The initial (100 hour) output of the HPS lamps is 15,000 lumens and of the CMH about 14,000 lumens. The difference in the lateral position of the peak luminance may be explained by the differences in lamp geometry and adjustment of the lamp position in the reflector. The same reflector, optimised for the smaller CMH lamp, was used for both lamps, with the lamp adjustment made before installation. TRL Limited 95 PPR043

106 Figure 4.30 Comparison of trial illumination under CMH 150W (top) and HPS 150W (bottom) A closer comparison for target position E presented in Figure 4.31 illustrates that this position had a much higher level of illumination under HPS. In particular, the luminance contours for HPS and CMH were different meaning that the target was in a relatively dark area under CMH and a brighter area under HPS. The position of the target next to the kerb may also have made detection more difficult, as the target merged into the kerb. TRL Limited 96 PPR043

107 Figure 4.31 Difference in luminance levels for HPS and CMH at target position E Observer effects It is also a possibility that the effects observed at position A and E are due to factors relating to the individual observers. As the largest difference was apparent for target position E this was chosen for further analysis. In order to keep the analysis simple, results were analysed in terms of the observers visual acuities or Snellen scores as these had been previously shown to be good discriminators for the data. Figure 4.32 and Figure 4.33 present a boxplot and 95% confidence interval information for target position E in terms of lamp and power setting for a particular Snellen score. TRL Limited 97 PPR043

108 200.0 Distance from target when observed (m) Lamp and Power CMH 150W CMH 95W HPS 150W 0.0 6/5 6/6 HPS 95W Snellen score Figure 4.32 Target E Boxplot Distance from target when observed (m) Lamp and Power 95% Confidence CMH 150W 95% Confidence CMH 95W 95% Confidence HPS 150W 95% Confidence 0.0 6/5 6/6 HPS 95W Snellen score Snellen score Figure % Confidence intervals of the mean for target position E TRL Limited 98 PPR043

109 These show that, for a Snellen score of 6/6, no significant difference could be determined for the mean detection distances. However, for observers with a Snellen score of 6/5 the difference can be seen to be significant at the 95% level. This suggests that for observers with normal (Snellen 6/6) visual acuity, no difference in detection distance was apparent between the different lamp types. Only observers with better than normal visual acuity (Snellen 6/5) demonstrated a difference related to lamp type for this target position. Further analysis of the data in terms of age range revealed a similar trend. Observers over the age of 60 years could not discern differences whereas those below 50 years could. This indicates that, for target position E, age and visual acuity factors affected the ability of observers to detect the target Effect of contrast ratio Contrast is known to be the key factor affecting the ability of an observer to see any given object, and may explain differences in recorded detection distances. For a flat target its contrast could be reasonably approximated by a function of the average luminance of the target relative to the average luminance of its background. Unfortunately the appropriate contrast parameter of a spherical target is much more difficult to determine, because there is a continuous gradient of luminance from the top to the bottom of the target. This results in continuous contrast gradient between the top and bottom of the target, and the contrast between the perimeter of the target and its adjacent background also varies continuously. Both flat and spherical targets may cast a shadow on the road. For a flat target, the size of this shadow will be dependent on its position in relation to the luminaires. A spherical target will always cast a shadow at its base, which will be larger than that for a flat target. The visibility of the target depends on an unknown function of the size and level of luminance of the areas, and the slopes of the luminance gradients. To investigate the target luminances and contrasts, images of the target were taken using the luminance camera, as described in Section , under each of the four lighting conditions and for each of the six target positions. For the analysis, semi-circular regions covering the top and bottom halves of the target were created. The luminances of these regions were measured using the software (see Figure 4.34), together with two adjacent semi-annuli which represented the corresponding top and bottom backgrounds. To allow more detailed future analysis, eight 3x3 pixel background regions, equally spaced around the periphery, were also included in the measurements (see Figure 4.35). Results of all these measurements are tabulated in Table I2 of Appendix I for the target, and in Table I3 of Appendix I and Table I4 of Appendix I for the background regions, under CMH and HPS lighting respectively. Table I2 also shows the average luminance of the top and bottom regions of the target taken together. A large number of possible contrast ratios could be calculated using this data. The contrast between the top and bottom halves of the target, calculated as ( L top Lbottom ) Lbottom, is tabulated in Table 4.13, together with the measured luminances of the top (L top ) and bottom (L bottom ) of the target and the corresponding background semi-annuli. It is evident that for target positions A, B, C and D calculated contrast ratios are reasonably consistent between CMH and HPS at both power levels. At position E and F a greater difference in contrast is apparent between CMH and HPS. On the basis that a difference in contrast between lighting conditions should correlate with a change in detection distance for a given target, target positions E and F should have both shown the greatest differences in detection distance between CMH and HPS illumination. However examination of Figure 4.29 show this is only the case for target position E. Also, for target position A the detection difference was significant, but the difference in this contrast ratio was small. The use of this parameter in further statistical analysis is therefore unlikely to fully explain the differences in target detection distance of positions A and E. TRL Limited 99 PPR043

110 Figure 4.34 Measurement region for target/background contrast measurement Figure 4.35 Regions used in contrast analysis TRL Limited 100 PPR043