Real-Time Yoga Pose Detection Desktop Application Using OpenPose and a Hybrid Machine – Deep Learning Framework with Enhanced Accuracy and Pose Coverage

This work introduces a novel method for evaluating various Yoga poses using advanced deep learning techniques. In our approach, a conventional PC camera is employed to capture real-time video, and multi-part pose detection is applied to identify the Yoga posture. Once a pose is detected, the system utilizes an improved scoring algorithm that is designed to provide consistent assessment across all types of poses. The proposed method supports self-guided Yoga practice by offering immediate feedback on pose accuracy, thereby assisting users in refining their technique. To validate the performance of our system, we conducted extensive experiments on a diverse set of Yoga poses across different environmental settings, which demonstrated its robustness and adaptability. Furthermore, we developed a hybrid machine learning framework that incorporates linear regression for extracting meaningful features from the key-points identified by OpenPose in each video frame. This integration not only enhances the recognition accuracy but also facilitates the real-time analysis of Yoga poses. Overall, our approach presents a comprehensive solution for Yoga pose assessment, paving the way for more effective and accessible self-learning Yoga applications.