Person Entry-Exit Tracking System

Developed a computer vision system that automatically detects and classifies person entry and exit events from CCTV video using object detection and appearance-based re-identification. The pipeline uses a YOLO-based person detector and a FusionReID (ResNet-50) model to extract robust body embeddings. When a person interacts with a predefined door zone, their appearance embedding is compared with embeddings from recent frames using cosine similarity to determine identity continuity and movement direction. This enables single-camera, online person re-identification for accurate entry/exit analytics while preserving identity privacy.