Real-Time feedback system for Accurate Yoga Pose Detection

The past couple of years have seen pose detection systems develop rapidly owing to their usefulness in enabling remote exercise wellness activities and reducing the reliance on physical instructors.In this work, we describe a new feature integrated into a yoga pose detection system: real-time feedback for automatic postural correction which greatly augment the interactions and performance precision with which people engage.Incorporating MediaPipe pose estimation together with a light-weight Convolutional Neural Network (CNN), the system achieves an average accuracy of 95.2% in detecting 15 common yoga poses.The newly integrated feedback mechanism calculates the skeletal keypoints for the feedback frame and in real-time provides corrective suggestions whenever the actual posture strays away from the expected set standards.The experiments conducted on the developed custom yoga dataset show that the system is usable and robust, particularly with regard to precision of pose realization, with novice users.This development improves the efficiency of implementing yoga remotely and provides new possibilities for safer fitness activities in the house.