Road Network Inference from GPS Traces using DTW Algorithm

This paper proposes a method for inferring the road network from Global Position System (GPS) traces, which is composed of intersections and the roads between each pair of directly-connected intersections. Random Sampling (RANSAC) algorithm is used to cluster the turning points, where the users change their moving directions, into intersections. All of the GPS traces are segmented by the intersections, resulting in connectivity matrix of the intersections and small GPS tracks for each pair of directly-connected intersections. At last, the road between each two directly-connected intersections is extracted through aligning and averaging all of the tracks using Dynamic Time Warping (DTW) algorithm. The main novelty of our methods is aligning the tracks point by point for each road using a "stretching and compression" strategy, which not only allows road estimation by averaging the aligned tracks, but also a deeper statistical analysis using their time alignment, such as analyzing the users' speed stability at a specific location. The experimental results show that our algorithm outperforms other methods by producing clean road network without spurious edges.