Turn match footage into a visual breakdown.
A tennis video analysis demo that tracks players and the ball, maps positions onto a mini court, and displays estimated movement and speed alongside annotated match footage.
Discuss a similar project ↗
See player positions, ball activity, and estimated movement in one view, instead of reviewing match footage without context.
The challenge
A standard match recording shows what happened, but extracting player positions and movement patterns takes careful manual review. The goal was to make those details easier to see while watching the footage.
The analysis workflow
The system annotates match footage, tracks both players and the ball, maps their positions onto a mini court, and displays estimated movement and speed alongside the video.
Three models, three jobs
A YOLOV26 model from Ultralytics detects and tracks players. A separate YOLOV26 model detects the ball. A ResNet50 keypoint model identifies 14 standard structural intersections and corner markers on the tennis court so the action can be mapped to a mini court.
Training data
The tennis dataset was sourced from Roboflow Universe. The demo brings the model outputs together in a visual analysis workflow for match footage.