AI Powered Yoga Trainer

Abstract

The Yoga is an ancient practice that enhances physical health, mental stability, and overall well-being through a series of structured poses and breathing techniques. However, without proper supervision, many practitioners struggle to maintain correct posture and body alignment, which can lead to reduced effectiveness or even physical strain. This project aims to address this limitation by developing an AI-powered Yoga Trainer that assists users in performing yoga poses accurately and safely through real-time posture correction and performance tracking. The system uses a webcam to capture live user movements and employs MediaPipe Pose estimation to detect thirty-three key body landmarks. Using these landmark coordinates, the system calculates major joint angles such as shoulders, elbows, hips, and knees, and compares them with predefined reference values representing the correct yoga poses. To ensure smooth and stable detection, an Exponential Moving Average (EMA) technique is applied to eliminate small fluctuations caused by camera jitter or minor body tremors. Based on the comparison, the system provides instant visual feedback by highlighting correct and incorrect joint positions, enabling users to adjust their postures immediately. In addition, the application tracks the duration for which each pose is held and evaluates overall accuracy, allowing users to monitor their progress over time. Developed as a web-based platform using Python, OpenCV, and MediaPipe, the system offers an interactive and user- friendly interface suitable for beginners as well as advanced practitioners. By combining artificial intelligence with fitness training, this project bridges the gap between traditional instructor-led sessions and digital learning, offering a safe, personalized, and efficient yoga practice experience that promotes continuous improvement, motivation, and overall wellness.