Asana Uplift: Elevating Yoga Practice with Deep Learning and Raspberry Pi 4

In the world of increasing health disorders, Yoga is the most effortless way to stay fit and healthy without experiencing useless expenditures. It is an ancient Indian practice that enhances overall mental focus, endurance, balance, and clarity. People can do yoga by hiring qualified & professional yoga trainers from studios or on their own with the aid of the Internet. Majority of the people prefer self-learning online because going to a trainer can be challenging for them in their busy lives. The problem with learning yoga on one's own is that one might not be able to recognise a wrong posture, which could lead to long-term chronic issues, pain and injuries. As technology advances, there are advancements in computer vision that aim to automatically recognize human actions from images and videos and give reports on improving postures. These advancements can also be applied to healthcare, security and sports. Human action surveillance can also be applied to recognizing a yoga pose and letting the user know whether it is accurate. In this paper, we proposed a simple model that tracks the movement of different body parts and measures the correctness of various yoga postures for the user. The proposed model also provides the correction feedback for yoga poses on pre-recorded and real-time videos. The accuracy of the proposed model is 93.43%. The paper also discusses various other deep learning models that have emerged in recent times. The proposed work aims on contributing to the emerging field of technology-enabled assistance systems, which explores the capabilities and constraints of key point extraction for human body landmarks in combination with deep learning.