Fine reconstruction of badminton swing dynamic trajectory assisted by event camera

The instantaneous velocity of smash actions in badminton can exceed 100 m per second, with action durations of only tens of milliseconds. Traditional frame-based cameras, limited by fixed sampling frequencies and finite dynamic ranges, struggle to achieve fine reconstruction of high-speed swing trajectories. This study proposes a fine reconstruction method for badminton swing dynamic trajectories assisted by event cameras, utilizing the microsecond-level temporal resolution and 120-decibel high dynamic range sensing characteristics of event cameras to overcome motion blur problems. The system adopts a heterogeneous visual sensor collaborative acquisition scheme, converting asynchronous event streams into tensor representations processable by deep learning through voxel grid encoding, and designs an event optical flow estimation network based on the contrast maximization principle to extract high temporal resolution motion features. The human pose estimation module obtains skeletal keypoint coordinates, and an extended Kalman filter framework fuses event optical flow, visual keypoints, and inertial measurement data to generate continuous three-dimensional trajectories. The fusion framework explicitly models observation uncertainty from each sensor modality, enabling robust state estimation under heterogeneous noise conditions. Experiments on a dataset containing 960 swing samples from 12 athletes validated the method's effectiveness, with trajectory reconstruction achieving a mean absolute error of 8.34 millimeters, a 42.3% reduction compared to traditional optical flow methods, and a processing frame rate of 82 FPS meeting real-time requirements. Ablation experiments show that the event optical flow module contributes 52.0% of the performance gain, and the multimodal fusion strategy further improves accuracy by 36.8%. Biomechanical analysis results reveal statistically significant differences between professional and amateur athletes in swing velocity curves, joint coordination, and other metrics, validating the system's application value in sports training guidance. A systematic error decomposition analysis and discussion of generalization considerations across diverse venues, lighting conditions, player styles, and equipment types are further provided to guide future practical deployment.