The Use of GPS in Travel Time Surveys

Transcription

1 The Use of GPS in Travel Time Surveys Rocco Zito, BEng(Hons) Research Engineer, Transport Systems Centre, School of Civil Engineering, University of South Australia. Michael A.P.Taylor, BEng(Hons I), M.Eng.Sc, Ph.D Professor of Civil Engineering, and Director, Transport Systems Centre, University of South Australia May Introduction Travel time surveys have long been used to provide performance data for the assessment of traffic systems. The traditional methods of finding the amount of travel time have been difficult to apply and sometimes provided only limited information. With the advent of the Global Positioning System (GPS) a new era in transport data collection has arrived. Not only will it benefit data collection in travel time surveys but other areas dealing with transportation issues will also gain, for example road asset management, Intelligent Vehicle Highway Systems, and real time vehicle tracking among others. Another recent breakthrough is the development of Geographical Information Systems (GIS). These systems allow data to be displayed in a spatial context as well as associating attributes to this spatial data. This concept has only recently become available on personal computers. The GPS-GIS combination forms an effective partnership in that the GPS system allows the efficient collection of data, while the GIS provides an interface that allows this data to be displayed together with its spatial attributes. Some GIS packages also perform queries on the database, with results thus being displayed graphically. Thus the combination provides a very powerful tool that can be used in the assessment of any transportation system. 2.0 Traditional Methods Traditional methods of performing travel time surveys can be classed into two categories, moving observer methods and stationary observer methods, as described in Taylor and Young (1988).

2 Moving observer methods enable quick and easy estimates of mean travel time. The basic resources required are, a vehicle, two people, and a data recording system. Data recording may be by paper, pencil and stopwatch, or by hand held or laptop computer, or by an instrumented vehicle. An instrumented vehicle is one that is fitted with a computer recording system such as the Travel Time Data Acquisition System (TTDAS) developed by ARRB. The School of Civil Engineering at the University of South Australia has a Toyota Camry sedan fitted with TTDAS. Two popular moving observer techniques are the floating car and chase car methods. Stationary observer methods collect travel time data by the use of teams of observers. situated at a series of fixed positions on the road network. Three available techniques are (1.) numberplate surveys, (2.) input output surveys, and (3.) path trace surveys. The major disadvantages of these methods are that they are very labour intensive and that they require skilled personnel to provide accurate and reliable results. As the speeds obtained by the use of this method are average speeds for the whole journey it is very difficult to obtain spot speeds or sub-section travel times by using this method. These methods also tend to be relatively expensive and so repeated tests are not common. An instrumented vehicle can provide detailed speed-time data, but such vehicles are expensive and uncommon. Results may be limited to the specific class of vehicle represented by the instrumented vehicle. On the other hand, the GPS unit is readily installed in different vehicles allowing detailed information to be collected for a wide range of vehicle types, for example trucks, buses and even motor cycles as well as passenger cars. Here is a considerable potential for detailed studies of the performance of a wide range of vehicles in urban traffic. 3.0 Advantages of GPS The Global Positioning System (GPS) is an American military based system that is designed to provide real time position, independent of weather, anywhere in the world. When the full complement of satellites is available the GPS system will provide full time positioning in all areas of the world. The position is determined by calculating the distances to a number of satellites encircling the earth and then performing a triangulation calculation to determine the position of the receiver. By utilising the GPS system many of the disadvantages of the traditional methods can be overcome. For travel time surveys only one person is required to drive the car, the data is recorded automatically. The accuracy of the GPS system is comparable to the accuracies gained by traditional methods given the correct planning procedures. GPS also has the advantage in that it can record data at rates of one reading per second.

3 A suitable GPS receiver can be purchased for around $2000 and a data recorder for about the same cost, however these costs continue to decrease. Once the hardware is purchased, the running cost are minimal, a tremendous advantage if many repeated tests are required. These are just some of the advantages the GPS system offer in travel time survey applications, as well as many other advantages in different areas of transportation planning, Lo Vecchio (1990). The following provides an account of an investigation of the use of GPS in a travel time survey. 4.0 The VTRACK System The School of Civil Engineering at the University of South Australia has purchased the VTRACK vehicle tracking system that uses the Global Positioning System to record in real time the position of a vehicle. The positional information can then be displayed in a Geographical Information System (GIS). The VTRACK system uses Mapinfo for DOS as its GIS. The VTRACK system uses Sony GPS receivers that have the capability of tracking 4 GPS satellites to acquire a positional fix on the vehicle. These receivers placed some limitations on the system since the accuracies they provide can sometimes be less than satisfactory for travel time survey work, without substantial data processing. Typical output from the VTRACK system can be seen in Fig 1. This data can be stored in the form of a dbase file or displayed directly in MapInfo for DOS. DAY MONTH YEAR HOUR MINUTE SECOND LAT LONG XCOORD YCOORD Fig 1 Typical VTRACK Output

4 VTRACK has the capability of displaying the position of the vehicle in real time at a central control centre. This is achieved by the use of a communications system between the vehicle and the control centre, however real time positioning was not utilised on the travel time runs performed due to the lack of a communication system. Instead, the data were logged on a laptop computer in the test vehicle. The School of Civil Engineering's instrumented Toyota Camry sedan was used as the test vehicle. 5.0 Data Processing When the travel time runs were performed the raw data was saved to a *.DBF file, and then processed using software developed for the purpose. This software allowed the data readings to be displayed in the MapInfo for Windows environment: this was chosen due to the Windows environment being more user friendly and having a greater range of capabilities for displaying and manipulating data, than MapInfo for DOS. Another feature of the software developed at the university is that with every data reading the VTRACK system produced a number of data attributes were associated to the reading, these include Individual ID number Longitude Latitude Time position was acquired Time from beginning of journey Total distance of journey to present point Distance travelled from last data point The speed at that point Stopped Time Bearing at that point These attributes were obtained by manipulating the raw data from VTRACK (Fig 1). The incremental time was obtained by subtracting the time at the start of the journey from the time of the point in question. The distance between two readings, given by VTRACK, was obtained by applying the following formula (MapInfo 1990). Distance (km) = cos (y y ) 2 2 (x 2 x1 ) + ( y 2 y1 ) 2... (1) where: x 1,y 1 are the longitude and latitude of the first point respectively in decimal degrees

5 x 2,y 2 are the longitude and latitude of the second point respectively in decimal degrees By using this formula on every point the incremental distance between each reading can be found, and hence the total distance travelled calculated. It must be noted that this formula is not exact, because it does not take into account the extra distance travelled due to the earth's curvature, but it does provide a very good approximation, especially for small areas. If the time between each reading is known as well as the distance travelled, the speed of the vehicle can be determined. Speed = Distance travelled / time taken to travel the distance...(2) Stop time was determined by dividing the amount of time the car has been stopped, (the criteria for a stopped car is explained in section 9.0), by the trip time at each reading, then expressed as a percentage. The bearing was obtained by finding the difference between the latitudes and longitudes of consecutive readings, the sign of the difference indicating in which quadrant the bearing lies. The bearing is then calculated using trigonometry, and associated with its appropriate heading. The bearings are broken down into 8 directions North (N), South (S), West (W), East (E), North East (NE), South East (SE), South West (SW), and North West (NW). An example of the typical results produced is shown in Fig 2, where the attributes for a typical point are displayed in a dialog box (MapInfo 1992). This run was performed on the Eastern Freeway in Melbourne on the 6 May 1993 during the AM peak. As is seen it is very easy to associate where the vehicle is by using GIS as well as knowing where it is in time by the attributes associated with each data point. In the past this type of association was very difficult to achieve: the implementation of the GPS system has made the acquisition of this data possible with a single person driving the car.

6 Fig 2 Travel Time Survey on Eastern Freeway In the Windows version of MapInfo there are two ways in which the data can be displayed. In the graphical mode the position of each data point is represented by solid circles as in Fig 2 ( Mapinfo for Windows allow may other different types of symbols to be used, besides solid circles). The browser format shows the raw data in text format with no graphical features (see Fig 3). The advantage of the browser format is that it allows the user to perform SQL queries on the database. The results of queries are highlighted in the browser format as well as in the graphical display. This is another powerful analysis tool that can be used in travel time surveys. With the graphical format the dialog box showing the relevant attributes can be displayed by first selecting the info tool, ι, in the Tools dialog box (see Fig 2) then clicking on the desired point. This allows for quick and easy access to the attributes of each individual point, which is very useful when analysing the journey. MapInfo for Windows has the advantage that the info tool can display attributes of the road network, such as road names, house numbers on the appropriate side of the road, and any other associated data. MapInfo also allows the road names to be displayed on the screen in many different styles and fonts thus adding to the presentation. Other

7 databases with relevant graphical information can also be displayed in Mapinfo, since it has layering capabilities. For example if postcode boundaries were important to the study they could be added as another layer. Fig 3 Browser Format of Survey Data This type of breakdown of the data also allows each segment of the journey to be analysed. For example the time taken to travel from intersection to intersection can be obtained, which could in this case be compared to the time taken to travel the length of the freeway. These comparisons could help a transportation planner make better decisions on congestion levels in specific areas. 6.0 VTRACK Errors The major disadvantage of the VTRACK system for travel time surveys is that it may not provide a suitable accuracy for the position of the vehicle. The error in the latitude and longitude of the car can be in the order of ± 50m. Under adverse conditions the error has been as high as 200m. Although this does not pose a problem in trying to determine the route taken for the journey, this error does pose problems when trying to determine distance travelled or speed at a given point. An example of the speed errors involved can be shown by a simple calculation. The VTRACK system logs data every 3 seconds: if there is an error of 50 metres in the position of the car, the speed error is then equal to 50m / 3secs 60 km/h. Obviously if the positional errors exist then the journey's distance is also in error. In all the tests performed it seems that VTRACK always over estimates the total distance travelled. This can be explained by the fact

8 that when the car is stopped an error of 50m is being added to the journey length every 3 seconds, even though the car is stationary. 7.0 TTDAS (Travel Time Data Acquisition System) TTDAS data were also recorded in the travel time runs on the Eastern Freeway. TTDAS is attached to the car's speedometer cable and gives accurate readings of distance travelled and speed second by second. The speed readings given by this system are accurate to within 1 km/h. This gives a basis on which to compare the results given by the VTRACK system. As can be seen in Fig 4 the distance travelled is over-estimated by VTRACK by about 4.4 km, however the slopes of the two graphs are similar. This indicates that VTRACK results could be used with some modifications to estimate the distance travelled. C o m p a riso n o f TTDA S & V TRA C K 4 M a y Distance (m) PM Peak TTDAS VTRACK Tim e (se c s) Fig 4 Comparison of TTDAS and VTRACK Total Distance Travelled Fig 5 shows the extreme variability of VTRACK speed when compared to the speed given by TTDAS. Again it should be noted that the trend of the VTRACK speed follows the trend of the TTDAS system relatively closely. Again this indicates that with some modifications the VTRACK speed values could be used as an approximation to actual speed travelled.

9 Comparison of TTDAS & VTRACK Speed Speed (km/h) TTDAS VTRACK Tim e (se c s) Fig 5 Comparison of TTDAS and VTRACK Speed (Raw Data) 8.0 Corrections 8.1 Distance Corrections that could be made to the raw GPS data to try and improve the total distance measurement include subtracting the distances given by GPS at the times the car was stationary from the total distance. This would be a tedious task since the readings at each intersection would have to be matched to the raw data and then subtracted. If total distance travelled is an important criterion in the travel time survey then the accuracy of this method would be questionable. A better way of determining the total distance travelled would be to use the ruler tool in MapInfo,, and then trace out the path taken by the car, hence giving the total distance travelled. This method will give more reliable results than the VTRACK system, but is dependant on the accuracy of the street map database: experience suggests that the street database of Mapinfo is suitable for most transportation studies. 8.2 Speed Fig 5 shows that the raw VTRACK speed measurements tend to over-estimate the actual speed given by TTDAS. For instance the maximum speed given by TTDAS for the Eastern freeway data set is 111 km/h whereas GPS gives a maximum speed of 257 km/h, which is physically impossible in a standard Toyota Camry! A technique that was employed in an attempt to get the GPS speeds to correlate with TTDAS speeds

10 involved ignoring all speeds greater than the maximum speed given by TTDAS. The results of this are shown in Fig 6. VTRACK Speeds < 111 km/ h TTDAS VTRACK 80 Speed (km/h) Tim e (se c s) Fig 6 VTRACK Speeds Less than 111 km/h As can be seen the fit of the line has improved, however there is still the problem that the GPS values still show a substantial amount of variability when compared to the TTDAS speeds. Again the trend of the GPS speed distribution closely follows the trend of TTDAS. This implies that if some sort of analytical procedure was applied to this new data a better correlation could be expected. A second technique that was applied to try and improve the fit was to apply a moving average to the new edited VTRACK data. An analysis was done using a moving average interval of 3, and the results are shown in Fig 7.

11 VTRACK Speeds with Moving Average Interval TTDAS VTRACK 80 Speed (km/h) Tim e (se c s) Fig 7 VTRACK Speeds with Moving Average Applied Fig 7 shows that the VTRACK line better matches the TTDAS line, however the speed values of the VTRACK data still differ significantly from those of the TTDAS line in certain regions of the graph. Moving average intervals greater than 3 were also tried however the actual speed values did not show much improvement. A reason for this could be due to the data readings from VTRACK being approximately 3 seconds apart, so averaging over 3 intervals averages, is effectively averaging the speeds of the last 9 seconds. The problem with a moving average approach is that prior values are used to approximate what is happening in the present. This concept is not practicable in vehicle tracking, since it is a dynamic system, continuously changing, in which the past has little to no effect on what is happening in the present. If the time interval between readings was reduced from 3 seconds to 1 second (or even less if possible) a moving average technique might be more appropriate in vehicle tracking applications. Unfortunately the VTRACK system has a fixed interval of three seconds with no option of variation. Another technique that could be employed to improve VTRACK speed and distance is the use of a Kalman filter, (SPRI 1993) a technique already widely used in the tracking of objects by radar. It dynamically fits models to what is happening in the present. For example if a right hand turn is made it will use the right hand turn model and ignore the other possible scenarios, until a change occurs and another scenario

12 better fits what is happening at the present time. Trials using this technique to try and improve VTRACK speed accuracy have not been attempted yet. The aim of future trials will be to assess the suitability of Kalman filters in vehicle tracking applications. 9.0 Stopped Time A useful property when trying to asses congestion levels during a run is to determine the amount of time the vehicle is stopped. As can be seen by the plot of VTRACK speed vs time the speed value almost never drops down to zero. However, as previously indicated the trend of the VTRACK speed follows relatively closely to the TTDAS speed. When the car has stopped the GPS speed does decrease, but not to zero. Therefore a threshold value was placed on the VTRACK speed. For example all VTRACK speeds less than 12 km/h represent the amount of time the car was stopped. This was found to serve as a good approximation to the actual stopped time. The results of setting this criteria are shown in Fig 8, where the VTRACK approximation differs by 1% from the TTDAS value. This criteria was used in a number of trials and the VTRACK stopped time was always within ± 2% of the TTDAS answer. Hence VTRACK can be used to give a good approximation on the amount of stopped time in a journey, provided the 12 km/h criteria is applied. Eastern Freeway 4 May PM Pe a k Stoppe TTDAS d Time 20% Eastern Freeway 4 May PM Peak VTRACK Stopped Time 21% Moving Time 80% Moving Time 79% Fig 8 Comparison of Stopped Time 10.0 GPS Receivers It must be noted that the receivers used by the VTRACK system are at the lower end of the market as far as GPS receivers are concerned. Their ability to only track the bare minimum of 4 satellites is a disadvantage, and is probably the major reason for the low accuracy of the system. With the advancement of GPS receiver technology the price of better quality receivers is coming down all the time, as is the trend with all new technology products. At the present time there are GPS receivers on the market

13 that are able to track up to 6 satellites simultaneously hence providing greater accuracy in positioning at a very competitive price. Tests performed with six channel receivers in vehicle tracking applications have produced accuracies that outperform those given by VTRACK. These receivers also provide more information to the user if desired, see Magnavox Corp (1990) and Trimble Navigation (1992). For example the speed and bearing of the vehicle are calculated by the GPS receiver and then recorded as output: these readings are more accurate than those calculated from VTRACK output. The user can also request the Position Dilution of Precision (PDOP) values, which give the user an indication of the integrity of the GPS position solution ie the lower the PDOP the more reliable the positional data. Height can also be obtained from these receivers if there are more than 3 satellites available: a feature not so important in travel time surveys but necessary if any kind of coordinate transformation is required. Another technique that is used in GPS surveying is to apply a differential correction to the coordinates given by the GPS receiver in the vehicle, as described by Nolan (1990) GPS Workshop (1992) and Hunter (1990). These correction terms are obtained by having another GPS receiver at a known reference point. Then the errors in position at the known point are passed to the vehicle and applied to the coordinates. Accuracies of less than ±5m are expected when this technique is employed. The many problems discussed in the previous section could be overcome by the use of better GPS receivers and the use of a differential correction therefore making GPS an alternative to traditional methods of travel time surveys. It must be stated however that the most recent versions of VTRACK have now incorporated 6 channel receivers as a part of the hardware The SVeeSix GPS Receiver The Trimble SVeeSix is a small "inexpensive" continuous tracking six channel receiver optimised for automotive applications. The primary output of the receiver is time tagged position and velocity at intervals of approximately one second. Other information is available such as PDOP as well as other related GPS information ref Trimble Navigation (1992). Since it is a six channel receiver it can obtain data from six satellites at any one time, therefore providing a better accuracy, and more reliability, especially when dealing with signal blockage, when compared to lesser channel receivers. A travel time run was performed using a SVeeSix, in the area surrounding the University of South Australia, Levels Campus, shown in Fig 9. As with the VTRACK

14 SV31R Fig 9 Travel Time run with SVeeSix tests the same Toyota Camry was used to obtain the speed profile and the total distance travelled. As can be seen in Fig 9 even at this large scale the vehicles path matches the centre line of the road relatively well. It should be noted that when the Melbourne travel time runs are mapped at this scale the vehicles path does tend to wander from the road centerlines, reinforcing the fact that a six channel receiver gives better positional accuracies than a four channel receiver. However it must be noted that the SVeeSix still overestimates the total distance travelled. Figure 10 shows that the total distance travelled is overestimated by approximately 7km, for a total journey length of 31km. As with VTRACK this can be explained by the fact that when the vehicle is stationary distances are still being added to the journey due to the errors in positioning. The big advantage with the SVeeSix is when the speed profiles of it and TTDAS are compared, Fig 11. It must be noted that the results shown in Fig 11 are raw results, no filtering or statistical techniques have been applied to this data. As can be seen the two profiles almost match each other. However due to the large scale of the graph it

15 seems Comparison of TTDAS and SV 6 Distance May SV 6 Distance (m) TTDAS Tim e (se c s) Fig 10 Comparison of TTDAS and SVeeSix Total Distance Travelled that the speeds match exactly for a large parts of the journey. Unfortunately this is not quite the case; when the speeds are time matched and subtracted from each other it was found that the average error was 0.9km/h with a standard deviation of 2.1 km/h. These are small errors when compared to the VTRACK system.

16 Comparison of SV6 and TTDAS Speed Speed (km/h) TTDAS SV6 31-May Tim e (se c s) Fig 11 Comparison of SVeeSix and TTDAS Speed It must also be remembered that the TTDAS system itself could be in error by up to 1km/h. The sort of accuracy given by the SVeeSix is quite adequate for travel time surveys, and again show how suitable the GPS system is for this type of work. Another interesting feature to come out of the tests with the SVeeSix is that the criterion for stopped time is that all speeds less than 1km/h are equivalent to a stationary vehicle. This value seems more realistic than the value of 12km/h used by the VTRACK system. As can be seen in Fig 12, the stopped times are within 1% of each other when this criterion was used, an accuracy comparable to any other system available. SV 6 31 May Speeds < 1 km/h Stopped Time 13% TTDA S 31 M a y Stopped Time 14% Moving Time 87% Moving Time 86% Fig 12 Comparison of Stopped Time

17 These results show that better accuracies can be obtained by the use of better equipment. It must be stated that the accuracies obtained by either the SVeeSix or the VTRACK system are comparable to those obtained by using traditional travel time survey methods. The major advantage of any kind of GPS system is that the amount of data that can be collected in a journey far exceeds that of traditional methods Disadvantages of GPS One major disadvantage of the GPS system is its inability to handle obstructions, Mc Lellan (1990): GPS will not cope with features like heavily tree-lined streets or tall buildings. The GPS signal sent from the satellites is just not powerful enough to get through the foliage. When this occurs the GPS system does one of two things: 1. No readings are recorded until the receiver has cleared the obstruction, then the receiver will regain lock on the available satellites and continue logging data. This usually occurs with large continuous obstructions such as treelined streets or large buildings extending for the length of the street, such as in the CBD. 2. The receiver will usually try and make a best guess as to where the car is. Usually this guess is not very reliable, so it can appear that the car has moved a vast distance in a short period of time. This could also be another reason for some of the very excessive speeds given by the GPS system. Errors of this type usually occur when a relatively small obstruction obscures the receiver for a short period of time. A good example of this are the overpasses on the Eastern Freeway in Melbourne, errors did not occur at all overpasses but did occur occasionally. However the filtering techniques, as described in previous sections, are usually able to handle both types of problems further supporting the use of the GPS system in travel time surveys. 13 Conclusions There is no doubt that the availability of the GPS navigation system is revolutionising the collection of data in transportation applications as in many other fields. Combine this with a GIS system that allows the effective display of all this data, and there is a transportation analysis tool, more powerful than any previously available. The ability of GPS and GIS to collect and display data for the entire journey as well as each individual section of the journey, is a concept that has never been practicable in travel time surveys. The minimal labour and hardware requirements of this system

18 must make it more attractive than traditional methods of transportation data collection and analysis. This paper has mentioned the limitations of the GPS system, but it has also dealt with the solutions to these problems. When looked at in a broad context, it is realised that the amount of detailed data collected by the system far outweighs anything that was available in the past. Issues such as accuracy and signal blockage when looked at in comparison with data obtained from traditional methods are relatively insignificant. The advantages of a GPS-GIS system far outweigh the disadvantages. It is hoped that these new technology tools if used properly will help transportation planners find better solutions to transportation problems, thereby raising living standards of the community, a desirable end that would certainly justify the means. References: Taylor, M.A.P. and Young, W, (1988),Traffic Analysis New technology & New Solutions, Hargreen Publishing Company. Nolan, J.M. (1990), Development of a Navigational System Utilising the GPS in a Real Time Differential Mode, University of New South Wales, School of Surveying. McLellan J.F,Scleppe J.B and Cannon, M.E,(Sept 1990), Performance of a GPS Based Automatic Vehicle Location System within Urban Canyons, GPS'90/SPG'90. LoVecchio J. and Van Dyke, K.(Sept 1990),Transportation Systems Applications of GPS, GPS'90/SPG'90. Hunter T. and Kosmalski, W. (1990), Vehicle Navigation using Differential GPS, IEEE. MapInfo Corporation, (1990), MapInfo User's Guide Version 1.1 MapInfo Corporation, (1992), MapInfo Desktop Mapping Software Reference, Version 2.0

19 Magnavox Advanced Products and Systems Company, (1990), MX 4200 GPS Receiver and MX 4200D Differential GPS Receiver Technical Reference Manual. Trimble Navigation, (1992), SVeeSix Boardset and Antenna Draft Specification. SPRI Estimation Theory - A Short Course,(1993),The Signal Processing Research Institute, Technology Park, Adelaide S.A GPS Workshop, (July 1992), University of South Australia, School of Surveying. Acknowledgments The authors would like to thank Geoff Rose, Russell Thompson and Elizabeth Ampt, of the Transport Research Centre, University of Melbourne, for their encouragement and contributions to the work carried out in Melbourne and also Mark Lethbridge of O'Halloran Hill, TAFE, for the use of the SVeeSix.