INTELLIGENT LAND VEHICLE NAVIGATION: INTEGRATING SPATIAL INFORMATION INTO THE NAVIGATION SOLUTION

Transcription

1 INTELLIGENT LAND VEHICLE NAVIGATION: INTEGRATING SPATIAL INFORMATION INTO THE NAVIGATION SOLUTION Stephen Scott-Young Dr Allison Kealy Dr Philip Collier Department of Geomatics, The University of Melbourne, Vic 3010 Tel Fax Key words: intelligent navigation, integrated systems, GPS, DR ABSTRACT Successful intelligent land vehicle navigation systems can only be realised through the integration of navigation data and spatial information. This is evident in the development of modern Intelligent Transportation Systems (ITS), where the Global Positioning System (GPS) is used to provide the navigation data, and spatial information contained within an information database is used to provide location details. With plans already underway for the development of a Global Navigation Satellite System (GNSS), the next generation of ITS will definitely incorporate satellite ing technologies. Unfortunately, the performance of any satellite technology is restricted in areas where sky visibility is completely or partially obstructed. There is a fundamental requirement to provide a robust navigation system to support future developments of ITS. Potential solutions include the development of integrated systems, which combine measurements from GPS and other complementary sensors, such as dead reckoning (DR), to improve the continuity of ing. However, current integration algorithms, such as Kalman filtering, have difficulty in contending with the high dynamics of land vehicles, and challenge the navigation capability of these systems within the environment of urban canyons. Ironically this is perhaps the one environment where the successful application of satellite technology could most benefit the ITS industry. This paper discusses the integration of the inherent intelligence of spatial information contained within a Geographical Information System (GIS) with measurements received from a navigation system. The spatial information provides additional data that is used to constrain the navigation solution and provide a more accurate and reliable estimate. With this approach, the solution is not dependent on the performance capabilities of the navigation sensors alone. It enables the use of lower accuracy navigation devices, thereby reducing the cost of navigation systems while still providing a viable solution.

2 INTRODUCTION Intelligent navigation is the process of improving the basic solution obtained from low cost navigation sensors for land mobile applications. This is achieved through the integration of measurements provided by the navigation instruments with additional spatial information contained within a map database. In the majority of current real-time vehicle navigation systems, a low cost GPS receiver is used to provide information on the vehicle s, and a Geographic Information System (GIS) is used to provide location details. For land vehicle navigation applications, GPS only systems are incapable of maintaining continuous navigation capability in environments where the satellite signals are obstructed (e.g. by buildings, trees etc). Solutions to this problem commonly involve the integration of GPS with dead reckoning (DR) sensors. This solution often increases the overall cost of the navigation system with little improvement in the solution, as DR systems suffer from the accumulation of errors over time. Additionally, complex Kalman filtering algorithms used for a more rigorous integration of GPS and DR measurements are often unable to cope with the high dynamics of land vehicle navigation. With the wealth of information contained in a GIS, data can easily be extracted and integrated into the vehicle navigation solution. In this way, apart from assuming a passive role of informing users about objects of interest in their surroundings, the information contained in the database is used as additional measurements within the navigation solution. This type of integration offers a solution that is capable of improving the accuracy and performance of low cost, low precision sensors for urban land vehicle navigation. DESIGNING A NAVIGATION SYSTEM The intelligent land vehicle navigation system developed for this research consists of both hardware and software components. The real-time navigation hardware component consists of: a low cost Garmin GPS receiver; a KVH fibre optic gyro (FOG); a Pentium 133, 64 megabytes laptop computer; an odometer. The software module developed in Smallworld Magik TM and Microsoft Visual Basic TM provides a user interface to the navigation software, a means of accessing the GIS database, as well as enabling intelligent navigation through the integration of measurements from the GIS with those from the real-time navigation system. The Hardware Components The system developed for this project is modular in its design. It therefore enables easy integration with various types of navigation instruments and techniques. Three modes of navigation are tested within this research: satellite navigation;

3 DR navigation; combined GPS/DR navigation. The satellite navigation mode relies solely on the GPS receiver. With the recent removal of selective availability (SA), the data obtained from the GPS receiver is accurate to ±12 meters 95% of the time (Hooper, 2000). The Garmin GPS 45 TM receiver used can track up to eight satellites simultaneously, supports the National Marine Electronics Association (NMEA) 0183 electrical interface and data protocol standard for communication between marine instrumentation, and has an RS-232 serial communication output (Garmin International, 1994). The specific NMEA sentences used by the navigation system were the Recommend Minimum Specific GPS/TRANSIT Data (RMC) and Global Positioning System Fix Data (GGA) sentences. The DR navigation mode utilises the change in the vehicle direction measurements from a KVH FOG and distance measurements from the vehicle s odometer. The FOG has an RS- 232 serial communication output at 9600 baud and is capable of measuring a maximum rotation rate of ±100 /second (KVH Industries, Inc., 1999). The FOG allows for input from a vehicle s odometer in the form of electrical pulses. Each pulse represents an amount of wheel rotation predetermined by the vehicle manufacturer. Data received from the odometer is converted into binary format and included with the information transmitted via the FOG s RS-232 output. This data is then used to compute distance travelled by the vehicle. The accuracy of the DR system is limited predominantly by distance measurement and is approximately 2% of the distance travelled. Since the DR system contains no means of absolute ing, navigation requires the provision of a starting location and direction. The combined navigation mode integrates both the GPS and DR sensors. In this research, because of the high relative ing accuracy offered by the DR sensors, the navigation system relies primarily on DR, resorting to the GPS navigation solution only when the difference between independently measured GPS and DR s agree to an expected level. To define this tolerance, the DR and GPS accuracies were taken into account. Given that the major error accumulation in DR measurements is from distance measurement and that the GPS measurement is accurate to ±12m 95% of the time, the difference between DR and GPS calculations should be within: ± (12m + 2% of distance travelled since last GPS measurement used) A flow diagram of the system hardware and data flow is depicted in Figure 1.

4 GPS Receiver NMEA sentences Laptop Odometer Wheel rotation pulses FOG DR binary data Figure 1 - Flow diagram of the system hardware and data flow Unlike the GPS receiver, the FOG does not constitute a low cost instrument. It was used initially to implement and refine the models for intelligent navigation. However, subsequent testing described in this paper will show that with intelligent navigation, such high accuracy devices are not required. The Software Component Implementation of the intelligent navigation system required a platform to provide a user interface to the navigation software, to facilitate the data integration between the different hardware, and to analyse and display spatial data. Smallworld 3 TM GIS was chosen for this purpose. Smallworld s open architecture and comprehensive spatial analysis functionality offered significant benefits in developing the software component of this project. The programming language of Smallworld 3 GIS is Magik TM, an object-oriented programming language that is also used to implement the majority of the core Smallworld 3 GIS product itself. Smallworld 3 GIS includes facilities for integrating applications programmed in languages other than Magik. This was a particular advantage as it enabled interpretation of the navigation device outputs to take place in Microsoft Visual Basic TM. While Magik could have been used instead, Microsoft Visual Basic contains comprehensive serial communication libraries that aided development in the communication between Smallworld 3 GIS and the GPS receiver and FOG. A flow diagram of the data through the navigation system software is shown in Figure 2. GPS NMEA sentences RS-232 connection Translation of NMEA sentences into the individual components of the RMC and GGA sentences by Visual Basic Satellite navigation data RS-232 connection DR binary data Translation of binary data into the individual DR components (change in direction and distance) by Visual Basic DR navigation data Smallworld GIS Figure 2 - Flow diagram of the data through the navigation system software

5 The User Interface The user interface for the intelligent navigation system was designed to minimise the amount of technical information supplied to the user, with its primary aim being simplicity of use. The user interface designed is shown in Figure 3. Position information Navigation device in use Number of satellites visible to GPS receiver Figure 3 The navigation system interface Navigation options, such as turning intelligent navigation on or off, automatically centering on the vehicle location and selection of navigation mode (i.e., GPS, DR or both) Accessing the Database Spatial data is a fundamental requirement for intelligent navigation. The road centreline data for metropolitan Melbourne was stored in the Smallworld 3 GIS database. This data can then be accessed via Magik, thus providing the essential link between navigation instrument data and spatial information. INTELLIGENT NAVIGATION Four principle rules of intelligent navigation have been identified in this research: closest road bearing matching access only distance in direction Closest Road The first step towards intelligent navigation is to make the assumption that the vehicle is travelling along a road (which is typically the case). This constraint can be included in the location solution, thus improving the accuracy of the computed of the vehicle. This simple algorithm is effective when the nearest road is in fact the road being travelled. However, when approaching intersections or when two roads are close to each other, the

6 ϕ (a) Measured (b) Actual Calculated Figure 4 - Correcting to the nearest road: (a) Navigation without correction. (b) Navigation with correction. (a) (b) Calculated Actual Figure 5 - Correcting to the nearest road with accumulated distance error ϕ: (a) Navigation without correction. (b) Navigation with correction. nearest road may not be the road being travelled. In these situations, searching for the nearest road downgrades the solution (Figure 4). Additional errors in DR navigation may arise. One such error occurs as the vehicle turns a corner. Due to accumulation of small distance errors, when turning a corner, the nearest road can still be the previous road of travel (Figure 5). Without the ability to determine absolute, further DR navigation becomes increasingly erroneous. Bearing Matching Clearly, as the closest road rule takes into account only absolute and not vehicle bearing, this rule alone is not sufficient. The second rule, bearing matching, requires that the nearest road to which the vehicle s is corrected must have a similar bearing to the direction of travel. This corrects the problems previously described. The threshold of similarity between the vehicle s bearing and the bearing of the surrounding roads may be adjusted to suit the accuracy of the navigation instruments. However, the larger the threshold, the more likely roads will be incorrectly matched as having the same bearing as that of the vehicle. The significance of this rule must not be overlooked when navigating using DR. Typically, the largest error source is introduced from distance measurements. As distances are dependent on wheel rotation, the odometer measurement is affected by tyre condition, pressure variation and vehicle speed (Madhukar et al., 1999). The combination of the closest road and bearing matching rules adjusts for this error each time the vehicle changes bearing above the threshold amount. For instance, the distance error ϕ, shown in Figure 5, is removed by intelligent navigation. The more often the vehicle turns a corner, the more frequently accumulated distance error is eliminated. Using DR as the only source of navigation over long periods of time, the accumulation of distance error may cause the navigation solution to become invalid. However, provided that regular change in direction occurs, as is often the case with city driving, accurate navigation by DR can continue.

7 Calculated (a) Access Only Figure 6 shows a case where application of the closest road and bearing matching rules incorrectly the vehicle. The access only rule is designed to identify and prevent this error from occurring. (b) Road B Road A Figure 7 - Road layout scenario Actual Figure 6 - Correcting to the nearest road taking road bearing into account: (a) Navigation without correction. (b) Navigation with correction. Road C Take, for example, a vehicle travelling along road A in the road layout diagram shown in Figure 7. Assuming the only route to road C is via road B, logic dictates that for the vehicle to be travelling along road C it must previously have travelled along road B. By logging previously travelled roads, the navigation system can prevent the vehicle from being located on a road that it could not possibly be on. Distance in direction This final rule further reduces the accumulation of distance error by calculating the distance travelled by the vehicle in the direction of the road rather than the direction measured by the navigation device. This is particularly important when navigation instruments of low accuracy are employed. For example, if a vehicle travels 1000m along a road of bearing 60 while measuring the road to have a bearing of 65 (i.e. 5 in error), an error in distance of 4m will occur (Figure 8). Although this may seem insignificant, over several kilometres, or with lower accuracy navigation instruments, larger errors can accumulate. This error is avoided by calculating the distance travelled independently from the bearing of the vehicle and then applying this distance in the direction of the road being travelled. IMPLEMENTING INTELLIGENT NAVIGATION 1000m 5 996m 1000m Figure 8 - Distance error propagated from bearing measurement error. The four rules of intelligent navigation were implemented using the Magik programming language. The fundamental requirement of the algorithm is the ability to search for roads (defined by centrelines in the GIS database) in the vehicle s vicinity (as determined by the navigation instruments). These road centrelines can then be interrogated for information such as distance to the uncorrected navigation solution and centreline bearing. The 4m

8 intelligent navigation rules are then applied to correct the solution. If more than one road matches all intelligent navigation constraints, the closest solution is selected. PERFORMANCE OF THE INTELLIGENT NAVIGATION SYSTEM The intelligent navigation rules were tested in two different environments, a suburban test circuit and an urban test circuit. The 5km suburban test environment was used to determine the performance of intelligent navigation without interference from external factors, such as satellite signal obstruction. The 3km urban environment was situated in the Melbourne central business district where GPS satellite visibility is severely restricted and provided proof of concept that an integrated navigation system with intelligent navigation in an urban environment could provide an accurate, continuous navigation system. On the suburban circuit, the different navigation systems of satellite and DR were tested independently. Figures 10 and 12 show the results of applying the four intelligent navigation constraints on a small part of the test circuit. For comparison, Figures 9 and 11 show the same part of the test circuit being travelled without intelligent navigation. The section of circuit shown in these figures is approximately one kilometre in length. Figure 9 - Satellite navigation without intelligent navigation Figure 10 - Satellite navigation with intelligent navigation Navigation began here Navigation stopped here Figure 11 - DR navigation without intelligent navigation Figure 12 - DR navigation with intelligent navigation

9 It is clear from Figures 9-12 that intelligent navigation is able to provide improved results. Figure 11 shows the accumulation of error in DR navigation. The start and end points of the navigation were in fact geographically the same. However, over the 1km section of suburban test circuit shown in Figures 9 to 12, errors of up to approximately 20m can accumulate in the DR system (Figure 11). This is reduced to less than 8m over the same distance when intelligent navigation is implemented (Figure 12). Further tests were conducted using a dual frequency GPS receiver system to provide accurate kinematic on the fly (KOF) s for measurement of the true vehicle trajectory. This test indicated that the mean RMS error between the intelligent navigation solution and the KOF solution was approximately 12m with a standard deviation of approximately 9m. Although this is not significantly different when compared with the raw GPS solution, which also had a mean of approximately 12m and a standard deviation of 9m, the advantages of intelligent navigation are apparent in Figure 13. Satellite Severe errors in GPS measurements possibly caused by multipathing DR Figure 13 Navigating the urban environment primarily relying on GPS Figure 13 depicts the results of navigation in the urban environment where GPS is primarily relied upon, only supplementing with DR measurements when insufficient satellite visibility occurs. During this navigation, urban canyoning caused frequent and prolonged periods of satellite outage up to 70% of the time. Additionally, multipath and deteriorating satellite geometry often compromised the precision of GPS measurements when signals were reacquired. These contributed to the subsequent errors in the navigation solution seen in Figure 13, where the DR system and the intelligent navigation algorithm are unable to correct for these errors. In Figure 14 this situation is reversed by using the DR and intelligent navigation as the primary navigation tools, only including GPS measurements in the navigation solution when they agree to the DR results to a specified level (as defined in

10 the section DESIGNING A NAVIGATION SYSTEM). The intelligent navigation system was able to detect when GPS measurements were in significant error and enabled 100% continuous navigation in the urban environment. Satellite DR Figure 14 - Continuous navigation in urban canyons The most significant impact of intelligent navigation is on the DR solution. While over the short term the amount of error correction is small, over longer periods of time intelligent navigation prevents the accumulation of errors to which DR navigation is prone. This enables sustained navigation in DR mode without requiring input from absolute ing devices. This factor is important for navigation within the urban environment where the ability to gain regular absolute s from GPS may not be possible due to obstructions. ERROR CAPABILITY TESTING An important aim of implementing intelligent navigation is to reduce the accuracy requirements of the navigation devices, thereby reducing the cost. Of the equipment required, only the FOG provides an issue in terms of cost. An alternative to the FOG would be to use a low cost digital magnetic compass. However, such compasses are restricted in accuracy by electromagnetic interference generated by the vehicle. In order to test the ability of the navigation system to cope with lower accuracy bearing measurements, an error was added to the FOG. A random error of ±30 was introduced to each measurement, thus allowing for a 60 window of error (Figure 15).

11 Figure window of error Figure 16 shows the results of introducing the 60 random error when travelling the same section of the suburban test circuit as in figures 9 to 12 without intelligent navigation. Figure 17 shows the result with intelligent navigation. Figure 16 - DR navigation without intelligent navigation and random error of ±30 Figure 17 - DR navigation with intelligent navigation and random error of ±30 Clearly, intelligent navigation was able to compensate for errors up to ±30. It is important to note, however, that all roads in this area intersected at approximately 90. If roads were to intersect at around 60, bearing errors greater than 30 could render intelligent navigation ineffective. However, with a high degree of error, limitations must be expected. Integration with other navigation devices (such as GPS) would enable errors to be avoided or corrected. CONCLUSION The integration of spatial information with measurements from low cost navigation sensors has proved highly successful in improving the continuity and accuracy of the navigation solution in urban environments. The most significant impact of intelligent navigation is on DR navigation. Without absolute capabilities, DR navigation is prone to the accumulation of errors that eventually render the solution meaningless. Intelligent navigation, however, largely eliminates this accumulation of errors, enabling sustained DR navigation without requiring input from absolute ing devices. This factor is

12 particularly important for navigation within the urban environment, as the intelligent navigation system is able to provide 100% continuity of the navigation solution. Intelligent navigation requires no additional equipment other than that already available in commercial in-car navigation systems, yet significantly reduces the accuracy requirements of navigation instruments. Hence lower cost instrumentation can be successfully implemented without compromising navigation performance. REFERENCES Garmin International, GPS 45 Personal Navigator TM Reference, Garmin International, U.S.A. Owner s Manual and Hooper, G., The End of SA, GIS User, Australia, Aug. Sept. 2000, 41, pp KVH Industries, Inc., KVH E Core 1000 Fibre Optic Gyro Technical Manual, KVH Industries, Inc., U.S.A. Madhukar, B. R., Nayak, R. A., Ray, J. K., Shenoy M. R., GPS-DR Integration Using Low Cost Sensors. ION GPS 99, Sept , Nashville, Tennessee, pp BIOGRAPHICAL NOTES Stephen Scott-Young is a final year Bachelor of Geomatics/Bachelor of Science (Computer Science) student at the Department of Geomatics, The University of Melbourne. His research interests include global ing, inertial navigation and geographical information systems and their integration. Dr Allison Kealy is currently a lecturer in the Department of Geomatics at the University of Melbourne, specialising in the research areas of GPS, GLONASS and integrated systems. Allison received her PhD in Geodesy from the University of Newcastle upon Tyne, UK in 1996, after which she spent 2 years in industry providing technical support for GPS/GLONASS manufacturers Ashtech Ltd. Dr Philip Collier is a Senior Research Fellow in the Department of Geomatics at the University of Melbourne. His research interests include; GPS deformation monitoring, dynamic least squares adjustment, and geoid modelling by least squares collocation.