Real Time Yoga Pose Recognition and Classification Using Movenet Deep Learning Model and Computer Vision
Shivabasamma Beli, Bhushana Patel, Mohana · 2025
In this paper implemented real-time yoga pose detection and classification using TensorFlow based MoveNet deep learning architecture and OpenCV. The application seamlessly integrates MoveNet to extract 17 key points and utilizes a neural network (NN) for the detection and classification of diverse yoga poses. Proposed NN model is meticulously trained on extensive dataset of yoga images, undergoing preprocessing to extract pertinent features through deep learning techniques. The embedded model on the website is adept at interfacing with a webcam or device camera, enabling users to observe themselves in real-time while executing various yoga poses. The model subsequently furnishes feedback on pose accuracy, derived from the neural network's classification outcomes. This proposed web application serves as a valuable tool for yoga enthusiasts seeking to enhance their pose precision. Additionally, it offers remote progress monitoring for instructors overseeing their learners' development. For each pose obtained correctness poses accuracy of${9 9. 9 7 \%}$. And for error postures obtained an error accuracy of${8 3. 9 2 \%}$.