Stereo Vision System based on the NVIDIA Jetson Nano for Real-time Evaluation of Yoga Poses

Yoga has become increasingly popular in recent years as it offers several benefits through physical, mental, and spiritual practices. Although online resources offer the possibility of self-training at home without the need for an instructor, unsupervised training can result in injury, as users are not provided suggestions on how to improve. In this work, a Stereo Vision System, hosted on an NVIDIA Jetson Nano, was developed to evaluate Yoga poses in real time. The MediaPipe Pose framework and Stereo Vision System were enabled to perform 3D angle estimation of several joints of the human body. These angles were used to design and implement a methodology to evaluate user performance and provide them with feedback. The proposed method provides a 0-100 score, as well as a color scale system to visually indicate how a user can improve as they are performing a pose. Results prove the Stereo Vision System can perform accurate 3D Pose Detection and Angle Estimation, and is, therefore, reliable for providing accurate and correct feedback in real time. The proposed system will aid in minimizing injury risk while improving physical and mental health of individuals that practice Yoga.