Real-Time Intelligent Surveillance with YOLOv8 for Threat Detection and DenseNet121 for Activity Recognition

The increasing complexity of global security threats necessitates advanced surveillance systems capable of real-time threat detection and forensic analysis. This paper presents an AI-driven surveillance framework integrating YOLOv8 for weapon detection and personnel classification, DenseNet121 for activity recognition module. The system features a scalable backend powered by Django and MySQL, while a Qt Designer and HTML and CSS-based frontend ensures seamless monitoring and automated alerting. Experimental results show robustness in challenging conditions such as low-light environments, and adaptability in various weather conditions. The weapon detection model achieved mean average precision at 50 percent threshold of 0.858, with a precision of 0.872. Personnel classification achieved mean average precision at 50 percent threshold of 0.947, with a precision of 0.946. Activity recognition exhibited reliable performance, classifying various threat activities in CCTV footages. This research addresses scalability, adaptability, and ethical compliance, contributing to the advancement of real-time surveillance technologies.