Abnormal Crowd Behavior Detection Based on Movement Trajectory

Abstract

Abnormal crowd behavior detection on surveillance videos plays an important role in social security and is an important research in computer vision. This work detects crowd abnormal behaviors according to motion trajectory. Specifically, the presented algorithm discovers abnormal direction and abnormal crossing of pedestrians, and uses a novel backtracking method to accurately mark the movement trajectory of abnormal behaviors in the crowd. Experiments were performed on the MOT16 challenge dataset to verify the performance of the proposed method. The experiment shows that the proposed method can effectively detect the abnormal direction, abnormal crossing and movement trajectory of pedestrian with the abnormal behavior.