Research on Human Motion Recognition in Sports Training Based on Deep Learning

Action recognition is a fundamental technology for health monitoring, human-computer interaction, robotics, and many other applications, and has always received widespread attention from researchers. In action recognition, it is very important to fully obtain and efficiently utilize feature information from multimodal action data, which is conducive to applying the model to practical life.The global average pooling layer and full connection layer are used to generate class labels with standard cross entropy loss. This article utilizes an extremely lightweight neural network model that combines skeleton joint distance features with three scale motion information to efficiently achieve skeleton based action recognition. The proposed model was experimentally compared on three datasets: SHREC, JHMDB, and NTU-RGBD. The experimental results proved the effectiveness of the model, with an accuracy of 80.2%.The human motion recognition technology in this paper is beneficial to the recognition of training movements and the effect of exercise.