AI-based solution for exercise posture correction

Abstract

In recent years, interest in all things related to physical wellness has been rising rapidly, especially after the COVID-19 pandemic – which brought about social distancing and many infrastructure shutdowns. Normally, people would go to the gym for wellness improvement, but during the pandemic lockdown, they can only do so at home with limited equipment and almost no direct guidance from an expert. However, recent achievements in human pose estimation have shown significant progress in the field, opening the possibility of a system that can utilize the real-time efficiency potential to offer posture correction. In this thesis, we present an human pose estimation-based exercise posture correction system leveraging a technique called batch processing and apply it to three weightlifting exercises. The system is able to correct practitioners’ posture in a real-time manner. The thesis work also includes web and Android applications to demonstrate the practical aspect of the system.