Human Re-Identification for Crowd Management

This project develops a smart crowd-monitoring system that counts each person only once, even if they leave and return, using human re-identification techniques. It integrates YOLOv8 for person detection, ByteTrack for multi-object tracking, and a ResNet-50-based re-identification model fine-tuned on a project-specific dataset. The system maintains a dynamic identity gallery that stores feature embeddings of individuals and matches them using feature similarity measures to ensure identity consistency across re-entries. A hybrid approach combining body features with face-assisted cues enhances the re-identification process. It is useful for event halls, public spaces, and busy buildings where knowing the actual number of people is important for safety and planning. By reducing repeated counting, the system provides a clearer picture of real pedestrian movement. This helps teams manage spaces more effectively, use resources wisely, and make safer decisions in crowded environments.