The 24-Type Tai Chi Classic Motion System Based on Kinect Motion Capture in Mobile Edge Computing

This study aims to develop an efficient 24-type Tai Chi motion recognition system by using Kinect and mobile edge computing (MEC) technology, to improve the teaching and training effect of Tai Chi. First, by recruiting Tai Chi experts and beginners to conduct experiments, a large number of 24-type Tai Chi movement data are collected to ensure the representativeness and generalization ability of the dataset. Second, in an experimental environment equipped with Kinect devices and MEC nodes, a Tai Chi action recognition model by utilizing Graph Convolutional Network with attention mechanism is constructed. The experimental results reveal that the proposed algorithm achieves 95.17% accuracy in action recognition, markedly higher than the prediction accuracy of the Convolutional Neural Network algorithm. In terms of data transmission, the proposed algorithm's average delay is stable at 0.53 seconds when the data volume is 6Mb, and the packet loss rate is maintained below 7.7%. The conclusion shows that the proposed algorithm not only has excellent performance in recognition accuracy and real-time performance but also has high stability and reliability in network data transmission performance, which provides effective technical support for Tai Chi's digital teaching and training.