Player Tracking and Re-Identification in Sports Video

A robust pipeline for tracking players in sports footage using YOLO for detection, DeepSORT for multi-object tracking, and a deep ReID model for appearance-based re-identification. This project is optimized for challenging, low-quality video and ensures consistent player IDs—even when players leave and re-enter the frame.

🚀 Features

• Player Detection: YOLOv8 model detects players in each frame.
• Tracking: DeepSORT assigns persistent IDs, maintaining player identity across frames.
• Deep Re-Identification: Integrates OSNet (Torchreid) for robust appearance-based matching, reducing ID switches and improving re-identification after occlusion or re-entry.
• Fallback Appearance Features: Uses color histograms if deep embeddings are unavailable.
• Ghost Track Reduction: Filters out short-lived and ghost tracks with max_age and minimum track age.
• Visualization: Draws bounding boxes, IDs, and (optionally) player trajectories.
• Optimized for Low-Quality Video: Tuned parameters and robust features handle missed detections and occlusions.

🛠️ Requirements

• Python 3.8+
• NVIDIA GPU (recommended for real-time performance)

Install All Dependencies

All required packages are listed in requirements.txt:

ultralytics
opencv-python
numpy
deep-sort-realtime
torch
torchvision
torchreid
gdown

Install everything with:

pip install -r requirements.txt

Tip: For GPU acceleration, install the CUDA-enabled version of PyTorch matching your GPU and driver. See PyTorch installation instructions.

📂 Usage

1. Download the YOLO Model Weights

Download the YOLO model weights (best.pt) from the following link:

Download best.pt

Place the file in your project directory before running the script. 2. Download the Input Video

Download the 15-second input video (15sec_input_720p.mp4) from the following link:

Download 15sec_input_720p.mp4

Place the file in your project directory before running the script. 3. Install all dependencies:

pip install -r requirements.txt

4. Run the tracking script:

python track_players.py

5. Press q to exit the video display window.

⚙️ Key Parameters

Parameter	Purpose	Typical Value
max_age	Frames to keep lost tracks before deletion	7–15 (lower for noisy video)
conf threshold	Minimum detection confidence for tracking	0.5–0.7
track.age	Minimum frames before displaying a track	3

🧩 Tips & Troubleshooting

• CUDA Errors: If you don't have a GPU or CUDA-enabled PyTorch, remove all .cuda() or .to(device) calls and run on CPU.
• Import Errors: If you see ModuleNotFoundError, install the missing package with pip install <package>.
• Performance: For real-time speed, use a GPU and consider a lighter ReID model.

🌟 Further Enhancements

• Save output video with OpenCV’s VideoWriter.
• Visualize player trajectories for movement analysis.
• Export tracking data to CSV for analytics.
• Fine-tune the ReID model on your own sports data for even better re-identification.
• Extend tracking to other classes (goalkeeper, referee, ball) by adjusting class filters.

📄 License

This project is for educational and research purposes. Please respect the licenses of the underlying models and libraries you use.

🤝 Credits

• YOLO model: Ultralytics
• Tracking: DeepSORT
• ReID: Torchreid
• Video processing: OpenCV

📬 Contact

For questions or improvements, open an issue or contact the project maintainer.

Ready to use! Fork, star, and contribute to make this tracker even better!

⁂