This project implements a Player Re-Identification system designed for sports video analysis. It automatically detects players in each frame, assigns them unique IDs, and maintains identity consistency—even when players leave and re-enter the scene. The system leverages a fine-tuned YOLOv11 model for detection and DeepSORT for real-time, appearance-aware tracking. The modular design allows individual component testing, feature extraction, and flexible customization.

• 🎯 Player Detection using a fine-tuned YOLOv11 model.
• 🔁 Player Tracking using DeepSORT for maintaining consistent IDs.
• 🧠 Feature Extraction including color, texture, and spatial features for re-identification.
• 📊 ID Management to match re-appearing players with previous IDs.
• 🖼️ Visualization with bounding boxes and track IDs.
• 🧪 Modular Testing of all components (detection, tracking, extraction, etc.).
• ✅ Easy setup via a single setup.py script.

player-reidentification/
│
├── data/                   # Input videos (e.g., 15sec\_input\_720p.mp4)
├── models/                 # YOLOv11 model weights (best.pt)
├── output/                 # Output videos with annotated tracking
├── src/                    # Core source modules
│   ├── detector.py             # Player detection (YOLOv11 wrapper)
│   ├── feature\_extractor.py   # Color, texture, spatial features
│   ├── tracker.py             # DeepSORT-based tracking
│   ├── visualizer.py          # Bounding box & track ID rendering
│   └── main.py                # Main pipeline runner
│
├── download\_model.py        # Script to download the YOLOv11 model
├── test\_system.py           # Unit tests for pipeline and modules
├── setup.py                 # Installation, setup, and testing script
├── requirements.txt         # Python dependencies
├── .gitignore               # Files and folders to ignore in Git
└── README.md                # Project overview and documentation

git clone https://github.com/priyam864/player-reidentification.git
cd player-reidentification

This will:

• Check Python version
• Create required directories
• Install dependencies
• Download YOLOv11 weights
• Run a basic test

python setup.py

📥 If the model fails to download automatically, you can download it manually:

Manual Model Link Save it as: models/best.pt

python -m src.main data/your_video.mp4

--output output/your_result.mp4       # Custom output path
--save-comparison                     # Save side-by-side comparison view
--show-detections                     # Show raw detections only (no tracking)

python -m src.main data/15sec_input_720p.mp4 --output output/tracked.mp4 --save-comparison

You can verify that all components (detector, tracker, extractor, visualizer) are working by running:

python test_system.py

This runs:

• Frame detection
• Feature extraction
• Dummy frame tracking
• Drawing visualizations

• The YOLOv11 model used is fine-tuned to detect players (class 0).
• The system is modular, allowing individual testing or replacement of parts.
• Videos are saved automatically in output/.

Dependencies are listed in requirements.txt and installed via setup.py. Major libraries include:

• ultralytics (YOLOv11)
• opencv-python
• torch, torchvision
• numpy, scikit-learn
• tqdm, requests, matplotlib, pillow, scipy

For questions or collaboration, feel free to reach out via GitHub.