GPS positioning using map-matching algorithms, drive restriction information and road network connectivity

Transcription

1 Extended abstract Submission for GISRUK 2001 GPS positioning using map-matching algorithms, drive restriction information and road network connectivity George Taylor 1, Jamie Uff 2 and Adil Al-Hamadani 2 1 School of Computing, University of Glamorgan, Pontypridd, Wales, CF37 1DL 2 Department of Geomatics, University of Newcastle, Newcastle upon Tyne, NE1 7RU Introduction A novel method of map matching using the Global Positioning System (GPS) has been developed which uses digital mapping and height data to augment point position computation (Taylor, 1999). Map matching algorithms attempt to locate a given vehicle position on a map. In a moving vehicle, this position may be assumed to be constrained to the road network (Bernstein et al., 1998). This novel method reduces the error in position provided by a low cost GPS receiver. This error is a sum from several sources, including signal delay due to the ionosphere and atmosphere, signal multipath and until recently selective availability (S/A). S/A was imposed by the U.S. military to purposefully degrade the accuracy of GPS, but was switched off on the 2nd of May 2000, and is to be replaced with regional denial capabilities in lieu of global degradation (Interagency GPS Executive Board, 2000). Since S/A error has been removed a low cost stand-alone GPS receiver will always provide a position within approximately 20m of the true position. However, the position error is often much lower than this. Taylor et al. (1999) describe this new map-matching algorithm, called the Road Reduction Filter (RRF) in detail. RRF is a method of detecting the correct road on which a vehicle is traveling along. The RRF achieves this by comparing road centre-lines and trajectories described by GPS point positions. Furthermore, the RRF includes a position error vector estimate using a formal least squares procedure, which may be calculated as the vehicle is moving (Taylor et al. 2000). This estimate can be thought of as a position correction from a virtual differential GPS (DGPS) base station, thus providing an autonomous alternative to DGPS for in-car navigation and fleet management. A quality metric called MDOP used in the RRF to place confidence bounds on rigorous decision making for the rejection of road centre-lines is described in Taylor et al. (2001). The RRF is implemented as an ArcView GIS application and is supported by a Dynamic Link Library (DLL) written in C++ using Borland Builder. ArcView is used to display all digital mapping and GPS point positions. It is extensively customized using ArcView s OO programming language Avenue. Customization consists of new user interface tools and ArcView functions that facilitate the interrogation of road center-line and DTM data, and linkage with the external DLL. A DLL written in C++ is required for computational efficiency. The DLL receives data from ArcView, reads GPS data in the form of RINEX files and performs almost all the required processing, including the computation of the GPS point position least squares estimate using satellite pseudoranges. Data and results are passed between the DLL and ArcView using external function calls. This paper describes further developments of the RRF map-matching algorithm. The algorithm is improved by the addition of a road network tracing routine and the use of intelligent road network data, which contains drive restriction information. A comparison between the existing RRF vehicle tracking algorithm performance and the algorithms performance enhanced with the new routines are reported. More detailed analysis and outputs of these comparisons will appear in the final paper as will the results of further tests and appropriate outputs. 1

2 RRF Map matching enhancement The improvement in map matching is achieved by the addition of two subroutines to the existing algorithm. A route network analysis routine, provided with ArcView s Network Analyst Extension, was incorporated into the existing software. This analysis software calculated the distance through the road network from a vehicle s previous position to each potential present position offered by the map-matching algorithm. Potential positions requiring an impossible network travel distance are instantly rejected using the newly incorporated network analysis software. This improved the reliability of the route tracking system at locations such as roundabouts or dual carriageways, and immediately after the loss of GPS point positions. A second routine was developed so that the RRF could interrogate Ordnance Survey (OS) OSCAR Traffic Manager with Drive Restriction Information (DRI), DRI is provided by ETAK on licensed to the OS. This routine will allow the algorithm to make decisions on road selection using road direction and access information. For example, on a dual carriageway each carriageway is coded as a separate one-way road. A roundabout is also coded as a one-way road, as of course is a conventional one-way road. A simple comparison between a vehicle s direction of travel (its bearing) and the bearing of permitted travel along a road segment is used for this test. M a p o f T e s t A r e a W i t h R o u t e s M a r k e d 2 Road Centre line Route 1 Route 2 1 N W E M i l e s S Figure 1: Map of the test area of Newcastle showing the two routes that were used. 2

3 Unfortunately, due to the high commercial value of the ETAK data, and the limited resources available, only OSCAR Traffic Manager without DRI was available for the project, kindly supplied by the OS. The data structure used with OSCAR for DRI records was obtained from the OS web pages (OS, 2000) and replicated. The required DRI was collected manually for all roads in the region of the two routes and used to populate the replicated OSCAR Traffic Manager data records. Before the development of these enhancements the RRF based its entire map matching on road geometry. The RRF now has intelligence to select roads by network distance traveled and drive restriction information. This enhanced RRF, incorporating two new routines, is called Intelligent RRF (IRRF). The two new routines were also tested independently of each other; these versions of the algorithm are called RRF-network and RRF-DRI. To evaluate the effectiveness of IRRF GPS C/A code observation data was collected in a vehicle driven on routes through urban Newcastle upon Tyne. Figure 1 shows the two routes that were used. Route one is a fairly well built up area and route two has good sky visibility. Over the same period, dual frequency phase data was collected in the vehicle and also by a static receiver recording base station data on the roof of the Department of Geomatics, University of Newcastle. This dual frequency data was used to compute a high precision (cm accuracy) GPS solution, which was assumed to be the true position of the vehicle at each epoch (one second). All available satellites visible to both receivers were used in the position solution computation, this number varied throughout the route from none to eight. Six different types of point position solutions were computed for each route, see table 1. RAW is the uncorrected stand alone positions, which is the same as that output directly by the receiver. Height aiding was applied in solutions 2 to 5 using OS Digital Terrain Model (DTM) data, obtained from the EDINA Digimap service (EDINA, 2000). Table 1. Six point position solutions Solution Type Data 1 RAW vehicle C/A code data 2 RRF vehicle C/A code data, RRF, digital road and height data. 3 RRF-Network vehicle C/A code data, RRF with network tracing, digital road and height data 4 RRF-DRI vehicle C/A code data, RRF with DRI data, digital road and height data 5 IRRF vehicle C/A code data, IRRF, digital road and height data. 6 RTK dual frequency phase data from both the vehicle and the base station Data processing All GPS data was transformed into RINEX format for processing by all methods, Gurtner (1993) provides a good description of the RINEX format. The processing of the data for solutions numbers 1 and 6 was carried out using the GPS post processing software GeoGenius. All other solutions were computed by processing the data with the various versions of the developed Road Reduction Filter algorithm. To correctly display the vehicle positions on OS large scale mapping, transformation software was developed as part of the work. The resultant latitude, longitude and height coordinates from the six solutions were transformed to OSGB36 (Ordnance Survey of Great Britain, 1936) National Grid, with a nominal transformation accuracy of 20cm (OS 1999). At each epoch where all six position solutions were available the difference in position between 6, the true position, and each of the other five solutions were calculated (Position Error). 3

4 Results On visual inspection, it could be seen that the original RRF selected the correct road segment when the vehicle was traveling along single carriageway roads. The largest proportion of road selection errors occurred where there were dual carriageways, slip roads and roundabouts. The RTK positions (assumed to be true) were good along all roads, as expected, but there were far fewer points. For both routes there was a total of 2039 points from the RRF calculation, but there was only 1060 RTK positions. This can be explained because to compute an RTK position at least five satellites are required, but because of height aiding from DTM data, all RRF solutions (2,3,4 and 5) can be computed with a minimum of three satellites visible. RRF-DRI The addition of the DRI routine has improved the road selection process. Route 2 in particular was improved unquestionably both through visual inspection and statistically. It can be seen on figure 2 that there were improvements often in the order of about 1.5 metres. Error is GPS positioning Error in GPS positioning on route 2 Error (m) Avg Time Without With Figure 2. Error in GPS positioning on route 2, with and without DRI processing RRF-Network RRF plus a network shortest path algorithm have improved reliability for positioning a moving vehicle. Table 2 displays the differences in Easting (DE), Northing (DN) and position (DP) between the RTK and RRF solutions for route 1, this is the error from the true position. Similarly table 3 displays these differences for the RTK and the RRF-Network solutions. It can be seen from the tables that there are small changes in the mean and standard deviation, with mean position improving by approximately 0.3m. However, removing gross errors, see figure 3, will further improve these figures. Further tuning of the RRF for network distance testing will possibly achieve this. 4

5 Table 2. Differences between RTK and RRF solution, route 1 DE DN DP Mean S. Deviation Table 3. Differences between RTK and RRF-Network solution, route 1 DE DN DP Mean S. Deviation Conclusions The main objectives of developing the two additional routines, one for network distance calculation and one to process Drive Restriction Information were to: prove the feasibility of adding intelligence to the road selection process. This was achieved. improve the accuracy of point positions provided by the RRF map-matching algorithm. This has also been achieved, but not significantly, yet. Position Error (RTK RRF-Network) Metres GPS Seconds Figure 3. Large position errors in route 1 Further work. There is still some work to be completed to fully realize a seamless integration of the new routines into the RRF algorithm. A full IRRF solution is still to be computed. The network distance comparison parameters have still 5

6 to be finalized to achieve maximum benefit from its inclusion. Further statistical outputs and graphs are required to present the results of the experiment. References Bernstein, D. and Kornhauser, A., (1998), Map matching for personal navigation assistants, 77 th Annual meeting, The Transport Research Board, Jan 11-15, Washington, D.C. Gurtner, W., (1993), RINEX: The Receiver Independent exchange format Version 2, Astronomical Institute, University of Berne. EDINA, (2000), accessed August Interagency GPS Executive Board (2000) Frequently Asked Questions About SA Termination accessed 12 December Ordnance Survey, (1999), Ordnance Survey datum transformation OSTN97 and geoid model OSGM91, OS, Southampton. Ordnance Survey, (2000), accessed 27 September Taylor, G., (1999), Improved real-time GPS positioning through the reduction of state space, Proceedings of GIS Research UK Conference, Southampton. Taylor, G. and Blewitt, G., (1999), Virtual differential GPS & road reduction filtering by map matching, in Proceedings of ION'99, Twelfth International Technical Meeting of the Satellite Division of the Institute of Navigation, Nashville, Tennessee, USA, pp Taylor, G., and Blewitt, G., (2000), Road Reduction Filtering using GPS, Proceedings of 3 rd AGILE Conference on Geographic Information Science, Helsinki, Finland, pp Taylor, G., Blewitt, G., Steup, D., Corbett, S. and Car, A., (2001), Road Reduction Filtering for GPS-GIS Navigation, Transactions in GIS, in press. Acknowledgements This work was supported by a Newcastle University Vacation Scholarship and a Nuffield Foundation Undergraduate Research Bursary (URB/00008/G). 6