Trajectory Clustering Using Affinity Propagation with Trajectory Entropy Descriptor

Abstract

To cluster similar trajectories in surveillance videos, weconsider using trajectory entropy descriptors, which arecomputed from trajectories of foreground objects. Based onthe descriptors, we can assess the distances betweentrajectories with different lengths and transfer the distancesto a similarity matrix for affinity propagation clustering. Inthe experiments, we have shown the effectiveness andefficiency of the proposed method for trajectory clusteringcompared to the state-of-the-art methods.