MINING SPATIAL DATA FROM GPS TRACES FOR AUTOMATIC ROAD NETWORK EXTRACTION

Abstract

The car manufacturing industry has been conducting a considerable effort to allow future vehicles to communicate, either between them or with a road infrastructure, in order to improve driving safety. As the position of each vehicle is an essential attribute of the proposed application protocols (to avoid collisions at blind intersections, for instance), and is also fundamental to support complex network protocols based on mobile wireless nodes with very limited transmission range, such communicating vehicles will be further equipped with GPS receivers. This massive distribution of GPS sensors, in conjunction with a free of charge communication infrastructure that allows accessing the information collected by such devices, will create a powerful new medium of remote sensing of geographical information. In this paper we address the automatic road network extraction based on this vehicular sensing infrastructure where the sensor in play is just the GPS receiver. We have resorted to the widely available GPS/GPRS tracking technology, heavily used by trucking companies, in order to obtain more than 30 million GPS points to construct the road map of an interesting city of Portugal, called Arganil, in an accurate, inexpensive and permanently up-to-date manner. Our algorithm is implemented using spatial SQL queries to aggregate data from multiple traces to produce a weighted-mean geometry of road axles, diluting GPS errors. In order to evaluate our extracted road network, we have compared its geometric and topological layers with a vectorial road map extracted from high resolution satellite images. Results show a highly accurate correspondence between them in all areas where a sufficient number of GPS traces have been collected.