Clustering Motion Trajectories Based on Isoperimetric Graph Partitioning Algorithm and Directional Trimmed Mean Distance

Moving object trajectory clustering is one precondition for object moving activity perception. Many researchers pay attention to clustering trajectory. In this paper, a novel method named directional trimmed mean distance (DTMD) is proposed at first to measure similarity according to analyzing some problems in existed method, such as no direction, yawp sensitivity, lower computation speed. Compared with other similarity measure methods, the value of DTMD is smaller between two similar trajectories and bigger between two dissimilar trajectories. It is simple Using this similarity, veracity is provided for clustering. In this paper, we also introduce spectral graph to cluster trajectories using DTMD distance as similarity in order to improve veracity for clustering. We do some experiments in real vehicle video scenes and then compare with other approaches such as LCSS and Hausdorff in the same scenes. The experimental results show that the validity and robust are both proved using our method.