Empirical Study of Deep Learning based Yoga Asana Identification and its Alternatives

This study focuses on the application of yoga, an ancient form of exercise that has been practiced for thousands of years. In recent times, yoga has gained popularity as a form of exercise and as a result, there is a growing demand for proper instruction. Many doctors recommended Yoga for the quick recovery of injuries as well as the best tool for combating mental-health issues such as depression, anxiety, and post-traumatic stress disorder. Yoga is the exercise that is widely accepted for its health benefits. Usually, there is a disconnect between what people are doing and what the ideal asana is. It is very important to perform yoga postures correctly to avoid variety of health problems such as joint pain, disc misalignment, shoulder pain, and so on. According to the researchers’ study report, nearly 87% of musculoskeletal pain or worsening injuries, and more than 10% said yoga caused pain in their hands, wrists, shoulders, and elbows. Our paper attempts to bridge the gap between what is the ideal posture and how to help people adopt a healthy lifestyle. So, in this research 12 Yoga asanas have been taken and classified them using neural network-based transfer learning models. Also, four transfer learning models named Baseline CNN Model, VGG16, VGG19, MobileNet and Xception models are applied. Out of all the models performed for Yoga asana classification MobileNet has leveraged a better accuracy of 91.2%. This study has employed deep learning methodology, and the proposed technique has achieved a high recognition rate. In addition, it was also discovered that the proposed technique has 80% more medicinal value.