Trajectory clustering based on length scale directive Hausdorff

Abstract

Trajectory clustering is the basis of scene understanding which helps interpretation of object behavior or event detection in video surveillance system. This paper proposes a length scale directive Hausdorff (LSD-Hausdorff) trajectory similarity measure. Firstly, the trajectory is encoded, and then we use proportion corresponding set, object position and its instantaneous velocity direction to represent the distance between two trajectories. After that, the hierarchical clustering algorithm is applied to cluster trajectories. In each cluster, trajectories that are spatially close have similar velocities of motion and represent one type of activity pattern. Finally, through experimental results in true scenes, we proved the accuracy and effectiveness of the proposed method in clustering.