YAMA: Real-Time Self-learning Yoga with Tensorflow Mediapipe

Yoga holds a wide range of physical and mental health benefits for the well-being of an individual, but what one needs to focus on is correct pose execution. It is crucial if an individual wants to reap the complete benefits of yoga and also wants to avoid injury. Traditional methods of pose correction greatly depend on human instructors, which was subjective and time-consuming. Yoga is an ancient exercise form originating in Indian society thousands of years ago, and has seen a modern re-emergence thanks to the integration of technology. The integration of these technologies enabled the creation of a comprehensive yoga pose detection system that can identify a wide range of yoga asanas with high accuracy. It also provides room for receiving instant feedback on the pose alignment and adjustment accordingly. In addition to that, this research work will also explore the possibility of adding audio or visual feedback mechanisms to enhance the user experience and facilitate self-guided practice. By capturing real-time data from the camera, individuals can receive instant feedback on their pose accuracy, facilitating self-learning and tracking progress towards their fitness goals. Online yoga is gaining popularity, offering lessons and tutorials at the click of a button. This integration of yoga with technology introduces a new dimension, where virtual AI yoga trainers come into play. These trainers utilize pose recognition tools like Mediapipe, leveraging ML5.js library for machine learning algorithms. Mediapipe offers direct access to pre-trained models for human pose detection within the browser. Our proposed model achieves a promising result, reaching <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 7 \%}$</tex>.