SecureVision: A Lightweight Real-Time Framework for Aggression and Unattended Object Detection in Surveillance Videos

The increasing prevalence of unattended baggage and aggressive behavior in public spaces poses significant risks to public safety, ranging from theft to potential terrorism threats. This paper presents SecureVision, a lightweight, real-time framework for detecting such anomalies in surveillance video. Leveraging YOLOv5 for object detection and BoT-SORT for multi-object tracking, along with handcrafted logic for threat assessment, SecureVision achieves high detection accuracy across classes including person, bag, and aggressive_posture. The system is deployed with a real-time web dashboard using Firebase and Kivy, and delivers inference speeds of approximately 30 FPS. Experimental results demonstrate that SecureVision is both efficient and deployable, offering a reliable solution for security monitoring in resource-constrained environments.