Computer vision and machine learning for real-time powerlifting form correction and personalized program adaptation

Abstract

Powerlifting demands not only physical strength but also precise technique to maximise performance and minimise injury risk. Traditional coaching often relies on subjective video reviews and standardised programs, limiting personalisation and timely feedback. This project presents an AI-powered coaching platform that combines computer vision and machine learning to analyse powerlifting form in real time and adapt training plans based on individual performance. Using a custom video dataset of 570 lifts (squat, bench press, deadlift), the system applies pose estimation and Random Forest regression to estimate Rate of Perceived Exertion (RPE), identify form deviations, and deliver corrective feedback. A web-based application allows athletes to upload videos and receive personalized insights. Initial user evaluations suggest improved technique awareness and high usability. This work offers a scalable, data-driven approach to powerlifting training that bridges the gap between subjective coaching and AI-enhanced feedback.