Matching of vehicle GPS traces with urban road networks

Precise positioning of moving vehicles is necessary for many applications but the accuracy of global positioning system (GPS) is seriously affected in dense urban areas. This paper presents a general framework for matching such vehicle GPS traces with urban road networks based on a pipeline approach, where the matching is divided into several sequential stages. The GPS position points with inadequate satellite geometry are first filtered by common filters of fix and horizontal dilution of precision (HDOP). Then three advanced filters are developed to filter the deviant position points that have good satellite geometry but are not consistent with the vehicle motion model, i.e. velocity, angular velocity and heading changes. Finally, two curve-to-curve distance measurement algorithms - Hausdorff and Frechet are implemented to match these filtered trajectories with urban road networks. The examination of this approach is carried out using the vehicle track logging data in Sugimotocho station area of Osaka city, Japan. The results show that the pipelining of several filter algorithms can remove error position points on trajectories significantly. It also suggests that the Frechet algorithm outperforms Hausdorff by including the topology of road networks.