Trajectory similarity measuring with grid-based DTW

Abstract

With the rapid accumulation of GPS trajectory data, the vast amount of spatiotemporal trajectory data hides extremely rich and valuable information with potential travel behavior patterns. The similarity of the driver’s travel trajectory is key to mining patterns, but how to reasonably and efficiently evaluate the similarity remains a challenge. To address this problem, we propose a driving trajectory similarity measurement using grid-based dynamic time warping (GDTW) to evaluate similarity of driving trajectory. Building trajectory grid vector model (TGVM), the method solves the problems of position shift and large computation for the similarity measuring of big trajectory data. Extensive experiments were conducted with a real trajectory dataset to evaluate feasibility and efficiency of the proposed approaches. The results show that GDTW performs a more robust and efficient processing of trajectory similarity than does traditional approaches, reducing by about 5 times of time-consuming.