Football & Fitness Video Analytics Pipeline

I have hours of CCTV footage from 5-a-side and 7-a-side turf matches and I want an end-to-end computer-vision pipeline that turns every recording into clear, per-player metrics. The system must treat fitness and football performance with equal weight. For fitness I care most about minutes played, distance covered and calories burned; for football performance the priorities are goals and shots. Anything else you can derive is welcome, but these numbers must be rock-solid. What I expect you to build • A repeatable pipeline that ingests raw CCTV video, detects and re-identifies each player, tracks their movement through the entire match and exports a CSV/JSON report plus simple visual overlays. • Fitness layer: automatic calculation of minutes on pitch, total distance, speed profile and calorie estimate per player. • Football layer: automatic event detection for goals and shots, linked back to the responsible player and time-stamped in the report. • A brief README that explains model choices, how to run the code locally (Python preferred), and how to retrain if the camera angle or lighting changes. Acceptance criteria • Works on at least two full, uncut sample matches I will supply (fixed overhead CCTV angle). • Player identification accuracy ≥90 % over the full match. • Distance error ≤10 % when compared with a manual benchmark. • Goal/shot detection precision ≥80 %, recall ≥75 %. • Outputs the required CSV/JSON files and an optional MP4 with overlays. You will likely lean on tools such as YOLO/Detectron for detection, DeepSORT/StrongSORT for tracking and standard football-model calibration for distance mapping, but I am open to any stack as long as the accuracy targets are met and the code is clean, documented and containerised. If you need extra metadata (team sheets, pitch dimensions) let me know early so I can collect it.