Fast Computation of Clustered Many-to-many Shortest Paths and Its Application to Map Matching

We examine the problem of computing shortest paths in a transportation network from a set of geographically clustered source nodes to a set of target nodes. Such many-to-many shortest path computations are an essential and computationally expensive part of many emerging applications that involve map matching of imprecise trajectories. Existing solutions to this problem mostly conform to the obvious approach of performing a single-source shortest path computation for each source node. We present an algorithm that exploits the clustered nature of the source nodes. Specifically, we rely on the observation that paths originating from a cluster of nodes typically exit the source region's boundary through a small number of nodes. Evaluations on a real road network show that the proposed algorithm provides a speed-up of several times over the conventional approach when the source nodes are densely clustered in a region. We also demonstrate that the presented technique is capable of substantially accelerating map matching of sparse and noisy trajectories.