Abstract:

Traditional video surveillance systems are known to be inefficient, they use vast amounts of storage by continuously recording all footage, even if it’s not relevant. This paper presents an intelligent surveillance system designed to address this problem through AI driven analysis and memory optimization. The system integrates a multi-stage AI analysis layer, using OpenCV for initial motion detection, YOLO (You Only Look Once) for real-time object and person detection, and MoViNet for efficient action recognition. A core memory optimization module ensures that only "flagged" clips of relevant events are stored, while non critical footage is either discarded or heavily compressed. The system also has metadata tagging for easy searching of data and a real time alert system (SMS/Email/Dashboard) for critical events. This event-triggered approach reduces storage requirements and provides a more cost-effective, scalable, and efficient security solution.