A Self Learning Yoga Monitoring System Based on Pose Estimation

Prahitha Movva, Hemanth Pasupuleti, Himangshu Sarma

Abstract: Modern life is stressful - long working hours, poor diets, inactivity, and increasing social isolation in the digital age have all contributed to rising rates of anxiety and depression. The COVID-19 pandemic escalated this situation with a series of quarantines. Without physical interaction, learning a new skill can be frustrating and stressful, especially a skill like Yoga that requires balance and physical coordination. On the one hand, many people cannot afford a personal trainer, but on the other, books and video tutorials do not offer personalized feedback. These limitations make learning Yoga on our own an overwhelming task. We propose a minimal investment model that will help people learn and practice correct Yoga forms from the comfort of their homes in an easy, stress-free manner - just by using their camera. A deep learning model is proposed based on PoseNet which gives an accuracy of 96.77% on the test dataset. We also conducted a survey to measure the satisfaction, efficiency and effectiveness of our system. Overall, 81.8% of the participants felt that our system helped them learn and perform the exercises better and 52.2% of them rated it as being “very good”.