VAD-Lite:A LightWeight Video Anomaly Detection Framework Based on Attention Module

Tracking anomalous activities from videos can enhance safety and early crime detection, thereby facilitating proactive measures in public security management and contributing significantly to the prevention of escalatory scenarios. By leveraging such advanced surveillance capabilities, stakeholders can swiftly respond to potential threats, ensuring a safer community environment and fostering a sense of security and well-being. In this article we propose a weakly supervised setting for video anomaly detection. The work explores the significance of attention mechanisms on anomaly detection by identifying key areas in a video frame. The proposed architecture is light weight; therefore, this can be deployed in edge devices to detect such activities in real-time. Our experiment is evaluated on UCF-Crime dataset and attained performance metrics comparable with those of other established models.