Design of dynamic ski motion capture and human posture detection model based on spatio-temporal Transformer and wireless device tracking technology

Abstract

As a result of significant advancements in living conditions, individuals have redirected their attention towards physical exercise. Skiing, as a widely popular sport, necessitates the real-time maintenance of correct posture during movement. Therefore, we present a dynamic skiing motion capture and human posture detection model that leverages wireless device tracking. Primarily, personnel tracking is enabled through the construction of service base stations and the utilization of wireless device tracking technology. Subsequently, a human posture detection model is formulated in the form of human posture key points, employing the image information of each frame obtained via wireless devices. Finally, we introduce a spatio-temporal Transformer structure that facilitates the detection and recognition of human posture in consecutive frames. Our results demonstrate that our approach can precisely locate and track the position of skiing personnel. Compared to the latest Blip and Conformer methods, our technique yields F values that surpass them by 1.20% and 4.51%, respectively. Moreover, our model can achieve convergent model parameters and accomplish training objectives more efficiently, thus enabling posture detection and dynamic capture of skiing personnel via image and video information.