Abnormal Crowd Behavior Detection Using Topological Methods

In this paper we present a novel framework for abnormal behavior detection in crowded scenes. For this purpose, the theory of topological simplification on the dense field is extended to the sparse particle motion field, which is used to describe the dynamics of the crowd. We propose two new methods for analysis of boundary point structure and extraction of critical point from the particle motion field. Both methods can be used to describe the global topological structure of the crowd motion, which is the kernel idea of our work. Various types of abnormal behaviors, including crowd formation/dispersal, crowds splitting/merging, can be detected by monitoring the changes of the topological structure. The advantage of our method is that each kind of abnormal event can be described as a specific topological structure change, therefore we do not need a complex classifier to detect these anomalies. Experiments are conducted on known datasets and results show that our method is effective in detecting and locating these kinds of abnormal behaviors.