A spatio-velocity model based semantic event detection algorithm for traffic surveillance video

Detection of vehicle events is a research hotspot in digital traffic. In this paper, an approach is proposed to detect vehicle events with semantic analysis of traffic surveillance video using spatio-velocity statistic models. The approach includes two successive phases: trajectory clustering and semantic events detection. For trajectory clustering, a statistic model of vehicle trajectories are presented, for which a spatio-velocity model is trained by analyzing the trajectories of moving vehicles in the scene. Based on the trajectory, which represents both the position of the vehicle and its instantaneous velocity, a trajectory similarity measure is proposed. Then, an improved hierarchical clustering algorithm is adopted to cluster the trajectories according to different spatial and velocity distributions. In each cluster, trajectories that are spatially close have similar velocities of motion and represent one type of activity pattern. For the semantic events detection phase, statistic models of semantic regions in the scene are generated by estimating the probability density and velocity distributions of each type of activity pattern. Finally, semantic events are detected by the proposed spatio-velocity statistic models. The paper also presents experiments using real video sequence to verify the effectiveness of the proposed method.