Enhanced Yoga Posture Detection using Deep Learning and Ensemble Modeling
Shah Imran, Zarif Sadman, Abidul Islam, Dewan Ziaul Karim · 2023
Yoga is one of the best at-home exercises for maintaining our physical health. However, yoga is all about successfully performing the 82 Yoga Asanas throughout the course of six classes. Lamentably, not everyone has the knowledge or can perform yoga accurately. So to do yoga poses correctly we will have to find a yoga instructor, but it can be very hard and expensive to find yoga instructors considering all possible general situations and status. Using Deep Learning(DL) and modifying some pre-trained models to some extent can be a possible solution to detect yoga pose and class separately, which can eventually help general people. This work proposes a detailed experiment using two pre-trained CNN models along with an ensemble model to detect yoga poses accurately. The work was done on a total of 18488 images divided into 6 major yoga classes and 82 different poses. The aftermath of using ensemble modeling was instrumental as it was able to detect yoga poses with a 95% chance of assurance.