Aerial movements in diving competitions significantly affect performance. Aerial motions are captured by video, but it is not realistic for divers to wear markers in competitions. In addition, markerless visual tracking of the human body and digitizing joint positions is extremely timeconsuming (Nomura & Goya, 2018). Recent years have seen progress in automatic objecttracking technology (Fujitake et al., 2021). Furthermore, Artificial Intelligence (AI) has enabled general-purpose pose estimation, which are OpenPose (Cao, 2017), Google PoseNet (Papandreou et al., 2018), and Vision Pose (Next-System, 2019). From the standpoint of scientific support for sports, the use of automatic tracking of divers and their pose estimation by AI is considered beneficial. However, the fact that the human body is not a rigid body and changes its shape makes automatic tracking difficult. Furthermore, since the pose estimation application was developed mainly assuming a posture in a standing position, it is not suitable for pose estimation in diving performances involving rotation of the sagittal plane. Furthermore, in competitions, officials and spectators may be reflected in the background, creating problems in tracking athletes and estimating poses. Therefore, this study aimed to improve the detection power of a diver's movement tracking and attitude estimation by devising image preprocessing.
© Copyright 2023 XIVth International Symposium on Biomechanics and Medicine in Swimming Proceedings. Veröffentlicht von evoletics Media. Alle Rechte vorbehalten.