Effective Trajectory Similarity Measure for Moving Objects in Real-World Scene

Trajectories of moving objects provide fruitful information for analyzing activities of the moving objects; therefore, numerous researches have tried to obtain semantic information from the trajectories by using clustering algorithms. In order to cluster the trajectories, similarity measure of the trajectories should be defined first. Most of existing methods have utilized dynamic programming (DP) based similarity measures to cope with different lengths of trajectories. However, DP based similarity measures do not have enough discriminative power to properly cluster trajectories from the real-world environment. In this paper, an effective trajectory similarity measure is proposed, and the proposed measure is based on the geographic and semantic similarities which have a same scale. Therefore, importance of the geographic and semantic information can be easily controlled by a weighted sum of the two similarities. Through experiments on a challenging real-world dataset, the proposed measure was proved to have a better discriminative power than the existing method.