Projects

Computer vision system that analyses football drill videos to measure player performance and generate structured assessment scores for coaches and scouts.

Highlights

• Built an end-to-end computer vision pipeline that converts football drill videos into structured performance metrics and a 0-100 execution score, using pose-derived features and drill-specific video analysis.
• Designed a human-in-the-loop camera calibration system with reusable geometry presets, perspective transformation, resolution-aware coordinate scaling, and validation checks to prevent silent measurement errors.
• Implemented asynchronous video inference and partial-result handling, allowing multiple drills to be processed independently while explicitly rejecting unreliable measurements instead of generating fabricated results; GPU inference runs at approximately 2x real-time.