Restricted-area violations, such as entering into vehicle zones, create serious security issues in a variety of monitoring applications. This research presents a deep learning-based framework for realtime detection and surveillance of individuals who violate designated restricted zones. The proposed system uses advanced object detection algorithms, specifically YOLOv8, for head detection and spatial reasoning to track individuals who enter restricted areas. The framework uses centroid-based tracking to accurately detect and count violations, ensuring that each individual is flagged once within a frame only once. The method improves detection accuracy further by modifying bounding boxes and using regionspecific polygonal filtering, allowing for more exact violation detection. Visual feedback is provided by overlaying boundary boxes and labels on the detected individuals, while cumulative violation counts are recorded. This method is highly effective, providing stable performance in changing conditions, and can be used for crowd management, security, and surveillance. The system’s architecture is flexible, with the ability to add capabilities like movement direction and speed analysis for more context-aware violations.

Computer Vision Traffic Surveillance YOLO Vehicle Speed Detection Direction Detection Helmet Detection Lane Violation Non-ANPR Cameras