Computervisionforfootball
One camera above the field — and the player gets a heatmap of passes, runs and pressures. The coach gets a full match report 10 minutes after the final whistle.
ProMatch — a platform for amateur and youth football leagues. Match stats used to come from human annotators a week after the game. We built a CV system: one camera above the field, and 10 minutes after the whistle there's a full report.
The problem
Manual annotation runs 8 hours per match. Players see the data a week later, when it no longer matters.
Constraint
No expensive hardware. One 4K camera above the field. An edge device next to it.
Goal
An automatic report 10 minutes after the whistle. Pass map, distance covered, pressures — per player.
From whistle to report
- 8 hours of manual annotation per match
- Stats arrive a week later
- 25% false positives on events
- Cost — same as for professional leagues
- 10 minutes from final whistle to PDF
- 94% event-detection accuracy
- All 22 players tracked across the whole match
- One 4K camera — no expensive hardware
How we built it
Detection
A YOLO model on 10,000 annotated frames. 22 players + ball + referee in frame.
Tracking
ByteTrack holds player IDs across the whole match. Losses under 3%.
Events
Event classifier: passes, shots, fouls. 94% accuracy on validation.
Report
10 minutes after the whistle — a PDF for the team and each player. Plus video highlights.
From video to stats
Match → frames → tracks → events → report. No human in the loop.
What came out
“A youth team got professional-grade match stats. Without a professional club's budget.”
Mark Reed
Yuka Kust
James Park
Vitaly Kust
Digital English teacher
A corporate AI teacher: speaks aloud, hears mistakes, remembers progress. HR gets a department-level report instead of chasing each employee.
