M.Brouwer, M.Kempe-CV at the Dutch Tennis Federation: Utilizing YOLO to create insights for coaches

Used YOLO computer vision models to create an affordable alternative to expensive Hawkeye systems for tennis match analysis

System performs court recognition, player detection/tracking, and rally detection through video analysis

Key challenges include:

• Varying camera angles and footage quality
• Dynamic camera movements
• Different court surfaces and lighting conditions
• Distinguishing players from ball kids/officials
• Working with limited federation budgets

Used color clustering and bounding box dimensions to identify and track specific players

Implemented coordinate transformation to convert screen coordinates to actual court positions

Built dashboards focused on key metrics coaches care about:

• Sprint distances
• Player positioning
• Match demands
• Rally analysis

Change management lessons:

• Start simple and gradually expand features
• Work closely with coaches to ensure buy-in
• Focus on metrics that directly benefit training
• Build trust through meeting specific coach needs/assumptions

System works best with high-quality broadcast footage but can handle lower quality recordings with some manual assistance

Solution provides cost-effective way for tennis federations to access advanced analytics previously only available through expensive systems

Focus on practical, actionable insights rather than complex analysis to ensure adoption by coaching staff