Wavelet-based Texture Model for Crowd Dynamic Analysis

Crowd event detection techniques aim at solving real-world surveillance problems, such as detecting crowd anomaly and tracking specific person in a highly dynamic crowd scene. In this paper, we proposed an innovate texture-based analysis method to model crowd dynamics and us it to distinguish the crowd behaviours. To describe complicated crowd scenes, homogeneous random features have been deployed in the research for behavioural template matching. Experiment results have shown that the anomaly appearing in crowd scenes can be effectively and efficiently identified by using the devised methods.