A video analytics pipeline for cricket that combines rally segmentation, player detection, pose estimation, and 3D animation to break down player biomechanics from raw match footage.
The clips below showcase the pipeline's output on sample match footage.
Player Detection
|
|
3D Animation
|
|
Simulation
|
|
At a high level, the pipeline processes raw match video through four stages:
- Rally Segmentation - isolates active rally footage from dead time / replays / crowd shots
- Player Detection - locates and tracks the bowler and batsman across frames
- Pose Estimation - extracts 2D keypoints for the tracked players
- 3D Animation - lifts 2D keypoints to 3D and renders an interactive animation for biomechanical analysis (e.g. bowling arm angle) Implementation details, model architecture, and code are not included in this repository - see the Results section above for output samples, or reach out below to discuss the approach in more depth.
If you'd like to discuss this project, potential collaborations, or licensing - feel free to reach out.
