From Athlete to Algorithm

In the presentation we demonstrate how computer vision and artificial intelligence applied to Olympic canoe athlete training videos can save significant amount of human resources.
Our system uses foreground-background separation to detect canoes and identify the waterline, then employs pose detection to locate paddle positions. We trained a neural networks to recognize biomechanically relevant paddle positions during routine training. The application of domain knowledge of paddle movements enhance accuracy in analysis. The solution significantly accelerates what traditionally required manual frame-by-frame video screening. Reducing a 20-minute per-athlete analysis to substantially less time.
Dr. Schuh brings expertise as a former Paralympic wheelchair sprinter who competed across three Games, achieving world championship status and European records in combination with a PhD in physics.