Deep learning head pose estimation for soccer head impact measurement from high speed video

Engineering of Sport 15 - Proceedings from the 15th International Conference on the Engineering of Sport (ISEA 2024)

Deep learning computer vision algorithms have successfully been adopted to automatically detect the occurrence of head impact in soccer in a large-scale broadcast video dataset. However, to achieve estimation of human kinematics directly from video, accurate player tracking is required. Deep learning human pose estimation has been researched extensively in the field of computer vision, and its application to full body kinematics measurement in sport has been proposed. The present work adopts a previously proposed markerless approach to angular head kinematics measurement in sport, which utilizes deep learning facial landmark detection algorithms to extract kinematics using head pose estimation (HPE).