Crowd Behavior Categorization using Live Stream based on Motion Vector Estimation

Abstract

The detection of anomalies in large crowd is a cognitive task. A proactive approach is required to effectively manage the crowd flow and to accurately detect the erratic behavior of crowd. In this paper, we present an algorithm which observes crowd optical flow in real time and detect any abnormal events in crowds automatically. The system takes the frames at regular intervals through a video camera and processes these frames using image processing techniques. The proposed system further uses certain rules to classify the normal or abnormal activities of crowd. We propose a novel motion vector based technique to detect behavior of the cluster of interest. The features of the motion vectors are analyzed to characterize the crowd behavior. The evaluation of the system is performed using different videos having different crowd behaviors and the results on simulated crowds demonstrate the effectiveness of the proposed system.

Keywords Motion vector, Crowd behavior, Real time processing