Computer Vision-Based Self-inflicted Violence Detection in High-Rise Environments Using Deep Learning

This article presents a system that utilizes computer vision techniques to identify suspicious behavior in high-risk areas. The technology employs cameras to detect individuals engaged in any suspicious activity or situated in hazardous locations. The system then analyzes the video using deep learning and computer vision techniques to identify those at risk. Upon detecting suspicious behavior or a self-inflicted violence attempt, the system promptly notifies the appropriate authorities, such as emergency medical services or law enforcement, enabling them to take preventive action. Videos are sequences of frames. We use a DeepSORT algorithm to extract the appearance and motion and connect different frames for real-time tracking with excellent performance and accuracy, even in crowded environments the information collected through video analysis can contribute to a better understanding of the underlying factors leading to such incidents.