Personal Training with Tai Chi: Classifying Movement using Mediapipe Pose Estimation and LSTM

This research aims to tackle challenges in the practice of Tai Chi Bafa Wubu (BWTC), where limited access to trained instructors and daily schedules hinder training consistency. The proposed approach combines Human Pose Estimation technology using Mediapipe with Long Short-Term Memory (LSTM) models to classify BWTC movements. This method utilizes video datasets collected from the internet and augmented to train LSTM models, focusing on An, Ji, and Zhou movements. Experimental results show that the model can predict movements with high accuracy in training and direct user trials. The development of these techniques facilitates more effective self-training in Tai Chi, leveraging advanced AI technology to improve movement supervision and user movement interpretation accuracy. This study not only offers a practical solution to enhance Tai Chi training efficiency and accessibility but also explores the potential application of pose estimation technology and machine learning in broader sports movement monitoring and evaluation. It is expected that this research will make a significant contribution to health and fitness by enabling individuals to independently practice Tai Chi with technological guidance, promoting better mental and physical health among the general public.