Social Force Model-Based Video Anomaly Detection Approaches

The research on spatiotemporal abnormal behavior recognition has two main categories, individual abnormal behavior recognition and crowd abnormal behavior recognition. Individual abnormal behavior recognition focuses on the classification of a single abnormal behavior or behavior trajectory that is the result produced by a single individual or a few persons. In crowd abnormal behavior recognition, much more individuals are involved and the abnormal behaviors cannot be detected as a simple collection of individuals but have to be regarded as a whole from the crowd all together. In this chapter, we introduce a novel method to detect and localize abnormal behaviors in crowd videos using the social force model which estimates the interaction forces among individuals in a crowd by treating individuals as moving particles. The interaction forces can then be mapped into image plane to obtain the force flow for every pixel in every frame and used to localize the regions of anomalies in the abnormal frames. The chapter ends with a case study for application where group panic and flee detection are analyzed by such technique.