A Real-Time Computer Vision Framework for Yoga Posture Analysis Using Coordinate Geometry

Abstract

Yoga is widely practiced without direct supervision from qualified instructors because of the growing popularity of at-home physical fitness activities. Unsupervised practice is convenient, but it frequently results in incorrect posture execution, which lowers efficacy and raises the danger of injury. The real-time vision-based posture analysis framework presented in this study is intended to support users during yoga practice by offering prompt feedback on posture accuracy. Key body joints are identified by the proposed system using MediaPipe-based skeletal landmark detection, and posture alignment is assessed using coordinate geometry-based angle computation. The framework uses lightweight geometric reasoning to enable effective real-time execution on common consumer devices, in contrast to data-intensive deep learning approaches that demand massive training datasets and significant processing resources. Through visual and audio signals, the system provides corrective feedback, assisting users in making dynamic posture adjustments. Usability and user engagement are further improved by a simple and interactive user interface. Experimental findings show that the proposed method improves posture accuracy and practice safety by successfully detecting posture variations and offering practical advice. The framework provides an approachable and dependable solution for unsupervised yoga instruction, which is especially helpful for learners and persons practicing on their own, including users who benefit from accessible, voice-guided assistance.