Surveillance Scene Analysis Model Based on Motion Trajectory
基于运动轨迹的监控场景分析模型

The proposed model learns the potential structure of motion trajectory within a hierachical process composing of classification and clustering. Topology prior was used to classify the spatial similarity of trajectory. After that by utilining Mixture Model to estimate distribution of motion features, the potential motion rules of the scene were obtained,based on which abnomal motion can be detected. The model is robust to low-level noises because of the statistic learning of multi-dimentional motion clues, and by combining prior the motion rules have obvious semantic structures. The real surveillance experiment results verified the efficiency of the proposed model.