Deep Learning-based System for Real-time Segmented Timing and Acceleration Analysis in Swimming.

Computer vision-based motion tracking has gained significant attention in various areas including sports. We developed a deep learning (DL)-based system combined with object tracking techniques to track swimmers’ information in swimming using real-time automatic segmented timing. The proposed system improved the efficiency of swimmer training and enabled coaches and athletes to evaluate their performance and adjust their practice methods. The system incorporated a YOLOv4-tiny tracking model to detect and track swimmers in real-time. This model was trained with approximately 100,000 images captured by underwater and above-water cameras. The system continuously tracked and marked the swimmers' positions using bounding boxes (Bbox). Segmented timing was determined by analyzing the time for the bounding box to cross predefined reference lines within the image. The system also used side-view image tracking, too to detect swimmer positions. Swimming data were displayed including segmented timing. The proposed system offered benefits for swimmers and coaches. Swimmers could monitor their progress, track their swim speed, and improve performance. Coaches evaluated swimming techniques, identified what to improve, and customized training plans based on real-time feedback and detailed analysis. The real-time automatic segmented timing and display of swimmers' tracking data provided a comprehensive overview of the performance and efficient training in swimming.