From Athlete to Algorithm: Transforming Canoe Technique Analysis with AI
summary
Pipeline automates frame selection and angle measurement for canoe training videos, cutting manual review from ~20 minutes to ~2 minutes per athlete and producing annotated video, CSV and debug frames. It combines densenet segmentation, PCA for alignment, open pose + hand pose for wrists and fingers, transfer learning on mobile net / yolo, and a custom paddle-state classifier trained on 35,000 labeled frames. System hits ~95% precision on test data, matches the coach frame exactly 70% of the time, and handles single-athlete footage robustly while multi-athlete cases still need work.
AI-written summary · about
chapters (11)
- 0:00AI in sports technique analysis
- 3:54Manual paddle pose analysis
- 9:37Computer vision fundamentals
- 13:30Aligning canoe videos
- 20:13Paddle pose detection
- 23:24Training pedal state classifiers
- 28:20Robustness with temporal filtering
- 34:0795% accuracy in paddle pose detection
- 35:52Analysis time reduction
- 37:17Future improvements and real-time feedback
- 43:07Q&A
related talks
36mFrom Athlete to Algorithm: Transforming Canoe Technique Analysis with AI
27mAnalysis of sports scenes using computer vision models
35mFootball video analysis using Deep Learning
40mYou can AI like an Expert
17mAI/ML demo derby
31mCV at the Dutch Tennis Federation: Utilizing YOLO to create insights for coaches
24mAI in maritime - Autonomous ships
17mFrom recovering Pro Golfer to AI Entrepreneur