A modular, lightweight multi-object tracking (MOT) framework combining YOLO detection, traditional tracking, Kalman prediction, and appearance-based Re-ID. Designed for real-time pedestrian and vehicle tracking under occlusion and identity switching scenarios.
| Feature | Description |
|---|---|
| YOLO Detector | Fast and accurate object detection using Ultralytics YOLO |
| Kalman Prediction | SORT-based tracking with Kalman filter for smooth trajectory prediction |
| Re-ID | Object re-identification (color/hog features) for occlusion recovery |
| Modular Design | Easy to toggle modules for ablation experiments via config.py |
- Clone the repository:
git clone https://github.com/yourname/multi-object-tracking.git cd multi-object-tracking - Install dependencies:
pip install -r requirements.txt
- Set your desired parameters and module switches in config.py.
- Just run!
python main.py
Regrading main objectives, key functionalities, detailed methods and test results, find in report.
The complete test data and results can be found in Google Drive.