The technology behind CourtEdge

Our in-house ML model, fine-tuned specifically on professional tennis footage across all surfaces. It locates the ball in every frame — even in blur, on break points, or when hidden behind a player.

↓ Custom CNN detector — trained on 2,000+ annotated trajectories

State-of-the-art detection model adapted to tennis, tracking both players frame by frame. Position, movement, court coverage — everything gets extracted and stored.

↓ Multi-object detection + tennis-specific tracking

We detect the court lines in each frame and reconstruct its exact geometry via homography. A Kalman filter smooths the output across time for stability — no jitter when the broadcast camera moves.

↓ Line detection + homography + Kalman temporal filter

Once the ball, the players, and the court are in real-world coordinates, we classify every point: serve direction, landing zone, rally length, pressure context… The raw data becomes tactical insight.

↓ Pattern matching + statistical tests + contextual tagging