Virtual Differential GPS & Road Reduction Filtering by Map Matching

A novel method of map matching using the Global Positioning System (GPS) has been developed for civilian use, which uses digital mapping data to infer the <100 meters systematic position errors which result largely from “selective availability” (S/A) imposed by the U.S. military. A method of rapidly detecting inappropriate road centrelines from the set of all possible road centre-lines that a vehicle may be travelling on has been developed. This is called the Road Reduction Filter. The S/A error vector is estimated in a formal least squares procedure as the vehicle is moving. 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. We derive a formula for “Mapped Dilution of Precision” (MDOP), defined as the theoretical ratio of position precision using virtual DGPS corrections to that using perfect DGPS corrections. This is shown to be purely a function of route geometry, and is computed for examples of basic road shapes. MDOP is favorable unless the route has less than a few degrees curvature for several kilometers. MDOP can thus provides an objective estimate of positioning precision to a vehicle driver. Precision estimates using MDOP are shown to agree well with “true” posiION GPS '99, 14-17 September 1999, Nashville, TN 1675 tioning errors determined using high precision (cm) GPS carrier phase techniques. INTRODUCTION The accurate location of a vehicle on a highway network model is fundamental to any in-car-navigation system, personal navigation assistant, fleet management system, National Mayday System (Carstensen, 1998) and many other applications that provide a current vehicle location, a digital map and perhaps directions or route guidance. Great many of these systems use the Global Positioning System (GPS) to initially determine the position of a vehicle. The Global Positioning System has become the most extensively used positioning and navigation tool in the world. GPS provides civilian users with an instant (realtime) absolute horizontal positional accuracy of approximately 100 meters. Most of this error is due to intentional dithering of the GPS timing signal by the US Department of Defense, an effect known as Selective Availability (S/A). This level of positional accuracy is insufficient to ensure that a vehicle’s location will correspond with the digitally mapped road on which the vehicle is travelling. A number of methods have been successfully developed to significantly improve GPS accuracy, the most notable being differential GPS (DGPS). Real-time DGPS can improve positional accuracy down to 1 to 5m. However, the use of real-time DGPS in a moving vehicle requires additional data in the form of pseudorange corrections (computed errors in the satellite range measurements). Continuous reception of terrestrial radio transmissions or communications satellite broadcast is required to receive these corrections. Often data can be combined from multiple sources integrating GPS with other navigational tools, such as attitude sensors such as the gyrocompass, vehicle odometer, flux gate compass and other dead reckoning methods. This use of multiple data sources again helps to correct for the error (noise) on the GPS position output. Multiple sensor data integration algorithms for vehicles are discussed by Mattos (1993). Dead reckoning produces the observed track by adding together the position vectors received from the sensor processor (Collier, 1990). The fact that vehicles are generally constrained to a finite network of roads provides computer algorithms with digital information that can be used to correlate the computed vehicle location with the road network. This is known as map-matching. Many methods have been devised for map-matching (Scott, 1994) (Mallet et al., 1995). Our research has developed and tested an algorithm that utilizes GPS for the initial vehicle position and geometric information, computed from the digital road network itself, as the only other source of data for mapmatching. MAP-MATCHING METHODOLOGIES Map-matching techniques vary from those using simple point data, integrated with optical gyro and velocity sensors (Kim, 1996), to those using more complex mathematical techniques such as Kalman Filters (Tanaka et al., 1990). A semideterministic map-matching algorithm, described by French (1997), assumes that the vehicle is always on a predefined route or road network. The algorithm determines where the vehicle is along a route or within the network by determining instantaneous direction of travel and cumulative distance. This is a dead reckoning system, driven by interrupts from differential odometer sensors installed on the left and right wheels. The system uses the digital road map to check for correct left or right turns and to remove distance measurement. The positional error is converted into along-track and crosstrack errors, allocating the first to the distance sensor and the second to the heading sensor errors (Mattos, 1993). For example, if the sensors indicate a 90 degree left turn and the digital mapping confirms this with the vehicle’s current position, the distance count may be reset to zero. Dead reckoning and map-matching systems like this are often linked with GPS receivers through software filtering schemes such as Kalman filtering (Levy, 1997). A mathematical framework for map-matching of vehicle positions using GPS is given by Scott (1994). The theoretical performance of a map-aided estimation process is assessed using error statistics to translate the raw positions onto the road network. However, Scott acknowledges that a key component of the map-aided estimator is correct road identification. All performance measures derived for the estimator are not applicable if the vehicle position has been projected onto the wrong road. This is true for performance measures of any map-matching algorithm. 1676 Systems that use only geometric information must utilize the “shape” of line segments (road center-lines) that define the road network (Bernstein et al., 1998). A logical first step is to determine which road center-lines are candidates for the vehicle’s true location. All road centerlines that cross the region of possible true position must be located, for example there are eight potential centerlines, highlighted in red, within the 80m region displayed in figure 1. This region will vary from a 100m radius circle around the computed raw/uncorrected GPS point position to a small error ellipse centered on the corrected position, with perhaps a semi-major and semi-minor axis True position Estimated position Figure 1. Potential road centre-lines of 5m and 3m respectively. The shortest Euclidean distance from the GPS position to each of these road segments is computed and ordered by distance. The method used here must first calculate A, B and C for the implicit and normalized equation of a line through two points that define the road center-line (line segment):