Tennis analysis, frame by frame.
A clean analysis layer for ball recovery, player movement, court projection, bounce detection, and hit attribution.
Computer vision pipeline
Tracks the tennis ball through tight windows and recovery scans, then keeps a confidence-aware trail for visual review.
Highlights bounce points and separates upper and lower player hits so event timing can be inspected against the video.
Maps video coordinates onto a normalized tennis court using detected field points, net posts, and calibrated homographies.
Builds smoothed upper and lower player tracks from ankle points, reducing flicker before measuring distance and density.
Lets the minimap and statistic panel grow from a selected start time as the video plays through a sequence.
Uses compact web data instead of massive CSVs, keeping the viewer responsive while preserving the original prediction context.
Current demo package
Interactive demo
The demo includes the video player, court minimap, field points, ball trail, hit and bounce events, live mode, running heatmaps, and time-range statistics.