Road network inference through multiple track alignment

Road networks are a critical aspect of both path optimization and route planning. This paper proposes to generate the road network automatically from GPS traces through jointly aligning tracks for each road segment. First, intersections are clustered from turning points where the road users' moving directions change. GPS traces are partitioned into small tracks for individual road segments by directly-connected intersections. The tracks for each road segment are aligned using a greedy method based on successor classification. A “forward-track” procedure is proposed to locate a warp path through jointly traversing all tracks in a way which keeps the points associated by the path element spatially close to each other. This involves an iterative procedure to cluster successor points on the tracks. The warp path produced during the alignment is used to average the tracks as the geometric representation of the road segment, and to analyze the velocity variation along the road segment. Experimental results show our method outperforms other existing methods in producing no spurious road edges and more accurate geometric road representation.