PoseRight: An AI Empowered Automatic Posture Estimation and Correction Model

Abstract

This paper introduces PoseRight, a novel framework that leverages real-time pose estimation and recognition technology to guide users in performing yoga and gym exercises correctly, thereby reducing the risk of injuries. PoseRight empowers users to achieve effective workouts without the need for gym memberships or personal trainers, addressing the growing demand for accessible fitness solutions. Through extensive experimentation with various deep learning models (CNNs, LSTMs, Transformers) and pose estimation techniques (MediaPipe BlazePose, MoveNet), the proposed framework achieves state-of-the-art performance, surpassing the accuracy of current research. PoseRight demonstrates exceptional test accuracies (99.01% for gym exercises, 96.8% for yoga exercises) while maintaining real-time processing speeds in both domains. This research signifies an advancement towards promoting safe and effective exercise guidance, ultimately fostering healthier lifestyles for users.