AI-based Pose Estimation System for Exercise Error Detection and Telerehabilitation Support in Resource- Constrained Settings

Nowadays, many people learn physical exercises from video tutorials on the internet, but tend to do exercises incorrectly by bending or stretching too much. This is more problematic in low-resource environments such as rural India, where access to physiotherapists is limited, and clinic visits may be costly. This paper describes the development of an AI-based system for estimating poses by tracking joints during home-based exercises with a single camera and providing an end-session report to both the individual and physiotherapists on errors in exercises and deviation from the standard poses. The AI-based system compares the poses of the individual with standard poses by tracking joints and estimates deviation in each repetition. A 2D pose estimation model has been proposed to test the feasibility of providing real-time support and summary analysis in telerehabilitation in low-resource environments.