Anomaly Detection in Crowded Scenes Based on Group Motion Features

Abstract

Event detection in crowded scenes is a challenging task for Computer Vision. In this study, based on group motion features, we propose an approach for crowded scene anomaly detection and localization. According to the motion trajectory of numerous pedestrians, both distance and relative speed between trajectories can be extracted, and the pedestrian groups can be recognized via their spatial relationship. Anomaly events in crowded scenes can be detected based on variations of group numbers and speed. To demonstrate the effectiveness of the approach, a quantitative experimental evaluation has been conducted on multiple, publicly available video sequences.