Labs / Computer vision experiment / 2026
Padel Vision
From pixels to court space.
A monocular match-analysis pipeline that tracks four players and the ball, projects play onto a court minimap, and renders annotated video.
- My contribution
- Computer vision research
- Built with
- YOLOv11 · PyTorch · OpenCV · Pose Estimation · Homography
Start with the work.
Broadcast footage provides one moving camera view, while useful match analysis needs stable player, ball, and court coordinates.
Adjie built a video pipeline combining fine-tuned ball detection, player pose tracking, court homography, and match overlays.
A moving camera. A stable reference.
- 01
Detects and tracks the ball at high resolution.
- 02
Tracks four players with pose estimation.
- 03
Projects detections through court homography.
- 04
Renders trajectories, identities, speed estimates, and match annotations.
Inspect the output.
Keep the uncertainty.
Player identities, ball tracking and the court inset are generated by the pipeline. Camera drift is registered against the first frame before projection. Hit markers require a ball turn within a player’s reach; some contacts remain unmarked.
Footage by UsaOne Ell, used under the Pexels License. This public demonstration uses licensed drone footage; development broadcast footage is not published.
The public record.
- Input
- Single broadcast camera
- Output
- Annotated video · court minimap