This repository demonstrates a well-structured computer vision pipeline focused on sports analytics, utilizing specialized trackers and heavy reliance on Pandas for data stabilization. * Leverages Pandas time-series analysis for robust smoothing and interpolation of noisy object tracking data. * The modular architecture clearly separates domain constants from specialized object and court line tracking components. * Effectively integrates state-of-the-art `ultralytics` YOLO models, forming an efficient computer vision tracking pipeline. * Improve the ball shot detection heuristic by converting rigid frame thresholds into configurable, physics-aware parameters. * The system uses persistent stubs (`.pkl`) to accelerate development iteration by bypassing repeated video inference.
Detailed description is only visible to project members.