Yoga Pose Classification Using Machine Learning

Yoga, an ancient practice rooted in Indian tradition, promotes physical fitness and mental tranquility. The COVID-19 pandemic has made attending a yoga class challenging, increasing the risk of injury when practicing without instructor guidance. To address this challenge, it was proposed to develop a system for evaluating yoga poses and determining their sequence. In this study, we developed a module that accurately identifies various yoga poses. Utilizing an open-source dataset of three distinct yoga poses, the system follows a two-phase development process. First, data points are extracted from images using the MediaPipe pose estimation library. Second, the obtained data undergoes preprocessing, followed by training and testing using classification-based machine learning algorithms. The algorithms employed include Logistic Regression, Random Forest Classifier and K-Nearest Neighbors (KNN). The module achieves an accuracy score of 96%.