Pose Estimation of Yoga Poses using ML Techniques

POSE ESTIMATION is a technique to identify joints in a human body from an image or video given as input to a computer. Pose estimation can be performed using Machine Learning (ML) techniques and Deep Learning techniques. Lately, it has been receiving lots of attention in the fields of Human Sensing and Artificial Intelligence. The main aim of pose estimation is to predict the poses of humans by locating key points like elbows, knees, wrists etc. In this work, we have proposed a model which uses Mediapipe, an ML framework, to obtain key point coordinates and ML algorithms like SVM, Gaussian Naive Bayes, Random Forest, Gradient Boost and K Neighbours classifier, which are compared and used to predict Yoga poses. Yoga is practised by people of all ages alike these days to fight issues caused both physically and mentally, thus improving the overall quality of life. Especially since the rise of the COVID-19 pandemic, the number of people practising yoga has only been increasing. In the model, human joint coordinates obtained are used as features. The model with the best accuracy and f score (MediaPipe+ SVM) is chosen for the final work. The yoga poses we used are Plank, Warrior 2, Downdog, Goddess, Tree and Cobra. On implementing the work, a real-time video feed from the webcam of the user's system is obtained, and pose estimation and classification of the yoga pose are done. Unlike in most current systems, suggestive measures to correct the yoga posture are also displayed in real-time alongside the webcam display of the person performing yoga along with some other basic pose information.