🧠 Real-Time Human Identification System

A real-time human detection and person re-identification system using YOLOv6 for object detection and OSNet (via Torchreid) for feature extraction and matching. Evaluated on public datasets with MOTChallenge metrics.

🚀 Overview

This project performs:

• Human detection in video streams using YOLOv6.
• Feature extraction using pretrained OSNet models.
• Human-to-human matching with cosine similarity.
• Real-time tracking and identity assignment across frames.

📦 Technologies Used

Purpose	Tools & Libraries
Object Detection	YOLOv6, PyTorch
Re-Identification (ReID)	Torchreid, OSNet
Real-Time Video Processing	OpenCV, imutils
Feature Similarity	Cosine Similarity (NumPy, Torch)
Evaluation	MOTChallenge (MOTA, IDF1 metrics)

📈 Achievements

• 🧠 Built a custom feature extractor using pretrained OSNet.
• 🧍 Detected humans in video using YOLOv6.
• 🔁 Matched identities using cosine similarity.
• 🎯 Achieved MOTA 63.8 and IDF1 62.4, ranked 95/100 on benchmark results.