A Survey on Investigating Machine Learning Approaches for Real-time Yoga Pose Rectification

Abstract

Yoga is an ancient technique, which is based on a science, which focuses on the harmony between mind and body. This practice is highly recommended by doctors for curing different health ailments, yet many struggle with proper execution of asana which may put the user at risk of injury. Hence, it brings the need to perform the asanas accurately. This paper delves into the various methods used to solve the difficulty of grasping yoga poses by precisely identifying and guiding practitioners in real-time and addresses these challenges by leveraging computer vision and machine learning (ML). The methodologies explored also include deep learning (DL), and hybrid models. Specifically, neural networks like CNNs and key point detection techniques, such as those implemented with OpenCV, OpenPose, and Mediapipe, significantly improve the accuracy of pose estimation. The integration of these technologies allows for real-time feedback, aiding practitioners in maintaining correct poses and reducing injury risks. Moreover, with a virtual yoga guide, users can practice yoga anytime also eliminating the hassle and expense of commuting to yoga centers and gyms. It helps the user maintain accurate yoga poses and avoid injuries which can hamper the body in the long term, making it easier to practice yoga at home. This makes wellness practices like yoga accessible and vision to have a huge contribution to a healthier society.