richYoga: An Interactive Yoga Recognition System Based on Rich Skeletal Joints
Yu-Hsuan Lo, Chun-Cheih Yang, Hsuan Ho, Shih-Wei Sun · 2021
In this demo paper, a yoga action recognition system is proposed. Based on the detected 3D positions of the rich skeletal joints provided by the face API, pose API, and hands API from MediaPipe, 543 skeletal joints are used train an LSTM neural network model to classify four yoga actions, including Half Wind Blown, Warrior II, Triangle, and Default. According to the generated yoga action classifier with 85% accuracy, a VR yoga tutor is implemented in the Unity environment to guide a user.