Yoga with Deep Learning: Linking Mind and Machine

Health and fitness play a crucial role in every aspect of an individual’s life. In an era where well-being is an absolute target, Yoga is one of the most straightforward ways to remain healthy and achieve mental focus, breath control and body balance without undergoing useless expenditures. The motive of this study is to combine the knowledge of Yoga with the continuously developing power of deep learning. In this study, we observe yoga pose estimation along with breath control and focus instructions for 10 different poses with the help of deep learning and Mediapipe. The proposed work predicts pose and instructions through model matching with landmark extraction done by obtaining keypoints on the pose by initializing the Mediapipe pose model. Here, a hybrid model is created using ANN with existing state-of-the-art models to achieve the pose estimation. The maximum accuracy achieved during the model implementation was around 93.43%. Later, when the model was deployed on Raspberry Pi-4 hardware, the maximum accuracy achieved was around 90.9%. The user also gets guidance on breath and focus control during a particular pose. Integration with deep learning and data science is seen to have created a whole new personalized Yoga experience with on-demand posture monitoring, customized feedback, and global access while creating future scopes for risk analysis and injury prevention.