Trajectory Analysis Method Based on Video Surveillance Anomaly Detection

Video surveillance system monitors potential malicious activities through video images, which is also an important part of city security prevention. In order to improve the application value of video surveillance system, how to change the mode of video surveillance system from post-forensics to active detection and prevention has become the focus of research. Because the suspect often paces back and forth and stays in the crime area abnormally, a large number of potential criminal behaviors can be excavated by extracting the track information of the target. However, video surveillance networks are widely deployed at present and massive data processing needs to consume a lot of manpower and material resources. In order to reduce the power consumption and resource consumption of the monitoring system, a hierarchical scheme based on cascaded RCNN multi-level head and shoulder Detection network (CRDNet) and trajectory analysis method is proposed to solve the abnormal event detection problem. This scheme only needs to analyze the trajectory information of pedestrians without establishing a sample data. Based on CityPersons, a mainstream pedestrian detection dataset, CRDNet achieved an accuracy of 0.73 and a recall rate of 0.65. Finally, this experiment is conducted on the UCF-CRIME dataset for theft events, and the identification accuracy is 81.9%. Experimental results show that the proposed scheme can effectively and timely detect the anomaly we are concerned about.