Yoga Pose Detection and Correction System

Abstract

This project presents a real-time Yoga Pose Detection
and Correction system built using HTML, CSS, and
JavaScript, leveraging the PoseNet model to detect
key human body points from webcam video. The
system uses a K-Nearest Neighbors (KNN) classifier
with ml5.js to identify the user’s yoga pose by
comparing detected keypoints against stored training
data. For pose correction, it calculates joint angles
and compares them with ideal reference angles to
detect deviations, providing instant visual and audio
feedback through p5.js and the Web Speech API. This
interactive feedback helps users adjust their posture
accurately in real time, promoting safer and more
effective yoga practice. By combining computer
vision, machine learning, and human-computer
interaction in a web environment, this project offers
an accessible and engaging tool for users to practice
yoga at home with guided corrections, enhancing the
overall quality and mindfulness of their sessions.