Human Pose Estimation for Smart Yoga using Artificial Intelligence

Abstract

People around the globe have been gradually realizing the importance of a person's physical and mental health. When people consult experts to improve their physical and mental health, they are constantly being responded to with a healthy diet, enough sleep, ways to reduce stress levels and many others, with exercises being the most common remedy. However, most of these exercises have a decent impact on physical health. Mental health is not much focused on, and that is where yoga comes to the rescue with a positive effect on a person's physical and mental health. For example, it helps boost immunity, relieves stress, improves brain functionality and body balance, etc. Practising yoga without any expert supervision can impose severe musculoskeletal injuries. Also, yoga studios or personal yoga coaching may or may not be affordable. With that in mind, we try to develop an AI-driven yoga coaching system that can track a person performing yoga and give feedback on the yoga poses done incorrectly by him/her in real-time. We used a pre-trained human pose estimation model called MoveNet to achieve this objective. We came up with an Artificial Neural Network based classification model YogANN that has been trained and tested on multiple datasets and achieves better accuracy on benchmark yoga-82 than previous state-of-the-art models.