Intelligent Yoga Coaching System Based on Posture Recognition

Nowadays, yoga as a popular sport has received more and more people’s attention. However, many people do not have enough time and suitable places to practice yoga. In recent years, the development of posture estimation methods has provided convenience for intelligent yoga coaches. This research aims to develop a yoga training system based on real-time pose estimation. The system is composed of a posture recognition network, a yoga standard movement posture library, a yoga movement correction algorithm and a system UI interface. Taking the real-time human motion video stream collected by the camera as input, the gesture recognition network based on OpenPose is used to extract the joint points of human gestures. According to our proposed correction algorithm based on entropy weight, real-time feedback and correction suggestions can be obtained. Experimental results show that this method can show high real-time performance on home computers and high accuracy on real data sets.