A Real-Time Anomaly Detection System for Video Surveillance Using DenseNet201
Suma S, Islapuram Lokesh · International Journal of Innovative Research in Information Security · 2025
Surveillance systems are pivotal for public safety; yet real-time anomaly detection remains a challenging task. Traditional systems rely on human monitoring, which is inefficient for large-scale deployments. This study addresses the critical rational for automated deviationsdetection by developing a system capable of identifying abnormal events in surveillance footage using a DenseNet model. The proposed solution captures live video input, processes it into minute-long segments, and classifies each segment as normal or anomalous. Anomalies include actions like fighting, arson, burglary, and shoplifting, and robbery, explosion, shooting, and stealing. To train the model, video datasets reformatted into frames, ensuring robust learning from diverse scenarios. If an anomaly is detected, the system sends alerts through email, phone calls, or SMS to notify relevant authorities, while normal footage is discarded to optimize storage. Initial outcomes demonstrate significant accuracy in detecting anomalies, emphasizing the DenseNet model's effectiveness. Beyond just reducing dependency on manual monitoring but also minimizes response times in critical situations. By automating the detection and alert mechanisms, the approach ensures enhanced security and efficient resource utilization. To wrap up, this study examines a scalable and efficient framework for real-time anomaly detection in surveillance footage, leveraging state-of-the-art deep learning techniques. Future enhancements will focus on multi-camera input integration and storing anomalous clips for forensic purposes, further advancing the system's applicability.