YOGGUIDE: DETECTING, RECOGNIZING AND TRACKING YOGA POSES USING VISION TECHNIQUES

Abstract

In computer vision, human pose estimation is a deep-rooted issue that in the past has revealed many challenges. In many fields such as security, video games, physical therapy, etc. analyzing human activities is beneficial. One ofthe challenges in human pose estimation is Yoga. These days, with stress and pressure full lives, people generally prefer doing yoga at homes as yoga is said to be art ofrelaxation, but they feel an instructor's need to evaluate their exercise form as doing wrong posture can health problems. Since these resources are not always available, human pose recognition can be used to create a system of self-training exercise that allows individuals to better learn and practice exercises by their own. This project objective is to develop an application which is an attempt to ensure correct yoga posture for three main poses which includes plank, warrior and pose reverse intuitive way. This project uses deep learning technique for pose estimation in which different stages are involved including pre-processing stage, data augmentation, creating CNN model and training the model. Yoga-guide's ultimate aim is to use pose recognition as a tool to allow a person to practice different poses ofyoga and classify the pose. In this project, using convolution neural network (CNN) model using Keras with TensorFlow as backend a deep learning model is proposed. The key benefit of using this technique is that it offers extraction and identification offeatures that are appropriate for pose recognition. We are able to achieve accuracy of 97%.