Design And Implementation Of Taichi Auxiliary Teaching System Based On Deep Learning

TaiChi has a long history and is a well-deserved treasure in traditional Chinese culture,and in recent years,the activities of TaiChi on campus have been in full swing.However,traditional TaiChi teaching lacks systematic teaching methods and scientific data analysis methods,which also leads to poor results in TaiChi teaching.In recent years,thanks to the gradual maturity of pose estimation technology in the field of deep learning,action recognition and action evaluation algorithms based on this have also become a hot research direction.Based on human posture estimation technology,this paper extracts the key points of three-dimensional bones in TaiChi movements,and proposes a TaiChi action recognition and evaluation algorithm on this basis,and finally designs and implements a set of TaiChi assisted teaching system,which serves students and teachers in TaiChi teaching,which can not only improve the quality of training,enhance students’ interest in training with AI applications,but also facilitate teachers to collect students’ practice and homework information,and empower traditional education with science and technology.The research content and related work of this paper include the following three points:(1)Research on TaiChi action evaluation algorithm.Based on the three-dimensional human pose estimation technology,this paper designs and proposes an action cosine similarity evaluation algorithm based on DTW algorithm for keyframe matching.The algorithm has a high correlation with the expert score on the self-made TaiChi evaluation dataset,which can help evaluate the completion quality of TaiChi movements.(2)Research on TaiChi action recognition algorithm.In this paper,an ST-GCN algorithm based on the new graph division strategy is proposed based on the original ST-GCN algorithm for the scenario of TaiChi movement,and the attention mask matrix is optimized to further strengthen the connection between limbs in the movement.Compared with the top1 recognition of the original ST-GCN algorithm on the NTU RGB+D dataset and the self-built TaiChi-Lite dataset,the top1 accuracy of the algorithm is improved.(3)Design and implementation of TaiChi assisted teaching system.This paper designs and implements a TaiChi assisted teaching system with action recognition and action evaluation algorithms in the field of deep learning as the core.Students can learn TaiChi through videos,submit TaiChi task assignments,evaluate their TaiChi movements,and view their TaiChi scores and task completion.Teachers can collect and summarize students’ task completion,visually analyze students’ mastery of each action,and publish homework tasks.The system has been tested to meet the various needs of users,meet the expected standards and can be applied in real scenarios.