Research on abnormal behavior detection technology in controlled areas based on human body parts trajectory tracking

Abstract During accelerator operation, beam loss can induce radioactivity in surrounding equipment, forming radiation hotspots that pose potential health risks to personnel entering the controlled area for maintenance. At the same time, the working conditions and behaviors of maintenance personnel directly impact the stability and safety of the accelerator system. Consequently, effective abnormal behavior monitoring within the controlled area is essential, enabling timely alerts and intervention to mitigate safety risks and protect personnel health. With the rapid advancement of computer vision technology, intelligent monitoring methods based on surveillance video have been widely applied in safety-critical environments. In this study, we propose an abnormal behavior detection framework that integrates person re-identification (ReID) with trajectory tracking. Specifically, the method first leverages surveillance footage to identify individuals entering the controlled area. Then, by incorporating a Human Body Parts detection model with the StrongSORT tracking algorithm, the system accurately detects and tracks key body components, including the head, body, hands, and feet. By analyzing the motion characteristics of these components and their spatiotemporal patterns, the proposed method effectively identifies critical abnormal behaviors such as unauthorized tailgating, door obstruction, prolonged presence in radiation hotspots, and accidental falls. Experimental results demonstrate that the proposed method achieves high recognition accuracy and real-time performance in laboratory tests, indicating its potential for deployment in real-world controlled environments. This approach enhances safety management within controlled areas, providing crucial technological support for intelligent monitoring in accelerators and other high-risk facilities.