Wrongly stretching your body or doing incorrect asanas, and breathing in an incorrect manner while exercising, can be detrimental to your health. Incorrect postures could lead to acute pain and long-standing chronic problems, say doctors. According to yoga injury statistics, yoga deteriorates 21% of existing injuries and triggers musculoskeletal pain in 10% of cases. 8% are willing to spend about $100 to get a memorable yoga experience.
Inspired by these facts, we decided to create an application to fix these problems and contribute towards an healthier and fit society. This ensures people can exercise with minimal support from instructors.
For ML part, The pretrained Model did not classify Horizontal exercises such as planks and Bhujangasana. To Overcome this challenge, we trained the Model with large amount of Horizontal Asanas and optimized the model. Providing Real time feedback was challenging as we had to monitor the asana continuously and comparing these results with our custom trained model was hard. We are providing custom and personalized live feedback to the user. We had to work extensively to come up with optimized algorithms, to find slopes of various skeletal segments as well as angle between the segments of important skeletal pairs during the particular exercise/yoga. Mapping Sets of skeletal nodes of user and custom trained dataset was challenging. We are mapping important Nodes of human body during Yoga, and using our efficient algorithms to give live feedback to the user so that he can correct his mistakes and inculcate healthy Yoga habits in his daily life.
Technologies used
Discussion