YOG-Assist
Saloni Bhambure, Saurabh Lawande, Rujuta Upasani, Jayshree M. Kundargi · 2021
Performing yoga requires training from the experts as it contains asanas which needs to be performed correctly for its best results. If a person doesn't have time for taking special training from the expert, then the person should be self-trained for understanding the correct pose. In this paper a model is proposed which can differentiate the right and wrong postures. The paper focuses on classifying the postures of Surya Namaskar asanas. Initially, by performing literature review we decided to use the OpenPose technique which is an open source pretrained model based on Caffe framework developed by CMV-Perceptual-Computing-Lab. Then dataset was created by capturing images of the people performing Surya Namaskar from Yoga Academy. The pre-trained model of OpenPose was used to extract the skeletons of the poses in images. Then a CNN model was built and trained on the above dataset. The validation and testing accuracy, and confusion matrix were found.