Marker-Less Athletic Posture Detection System Based on Human Pose Estimation via Multi Cameras for Bowling

This paper presents a sensorless athletic posture detection system for bowling by utilizing multi-view human pose estimation with YOLO v7. Firstly, audio synchronization is used to ensure the accurate alignment of the captured data. Afterward, the human joint locations from the image being taken by multiple cameras are detected. And then, a three-dimensional (3D) posture is generated by synthesizing both joint locations. This approach eliminates the need for markers, addressing the challenge of acquiring accurate posture data without impeding gameplay. Experimental results show that the system demonstrates performance in accurately estimating body joint positions and angles, enabling precise analysis of key aspects such as balance and arm swing.