Semantic Understanding of 3D Shuttlecock Trajectory

During a badminton match, each rally—from the serve, through successive returns, to the dead ball—constitutes the shot-by-shot rally structure, which forms the temporal basis for structured shot-by-shot annotation. In net sports, the ball’s flight trajectory provides data about the temporal organization of play and ball-control performance. This study performs semantic analysis of 3D shuttlecock trajectories to detect structural events in a match and to extract players’ control-performance metrics. First, time-synchronized multi-view 2D localizations from a camera platform are fused to reconstruct 3D trajectories. Next, semantic analysis of the 3D trajectories detects the structural events of serve, return, and dead ball in badminton matches. Finally, shot-level control-performance metrics (e.g., shuttle speed, elevation angle, and direction) and technical metrics (e.g., stroke type, over-net height, and landing point) are derived. These capabilities are collectively referred to as a “semantic understanding of shuttlecock trajectories.” On the data processing side, we propose filtering techniques targeting erroneous 2D detections to improve the accuracy of 3D reconstruction. We further design a trajectory-fitting algorithm based on a shuttlecock flight model to correct and compensate for errors and misses from deep learning–based object detection, and to predict out-of-view trajectories and events. This research constitutes a core component of the Content Layer within a smart sports spatial sensing architecture and can benefit the development of intelligent applications.