Title: Study on athlete's human motion recognition method based on deep attention mechanism

Authors: Yuntao Chang

Addresses: Department of Physical Education, Hunan City University, Yiyang, Hunan, 413000, China

Abstract: A method of athlete's human motion recognition based on deep attention mechanism is proposed in this paper. Firstly, the Kalman filter is used to denoise the action data. Then, the depth attention mechanism is introduced to design the extraction method of the angle between the bone length and the bone, extract the depth space feature, and use the attention mechanism to aggregate the depth space feature to complete the athlete's human movement feature representation. Finally, combining the batch-hard triple loss function and the softmax cross entropy loss function, the human motion recognition function is trained to realise the athlete's human motion recognition. The experimental results show that when the running time is 20 minutes, the recognition recall rate of this method is 98%, the recognition accuracy is 96%, and the recognition time is 9.0 seconds. It shows that this method can effectively improve the recognition effect of athletes' body movements.

Keywords: deep attention mechanism; Kinect DK body camera; loss function; spatial characteristic matrix.

DOI: 10.1504/IJRIS.2025.145077

International Journal of Reasoning-based Intelligent Systems, 2025 Vol.17 No.1, pp.58 - 65

Received: 14 Mar 2023
Accepted: 27 Apr 2023
Published online: 18 Mar 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article