Real-Time AI Posture Correction for Powerlifting Exercises Using YOLOv5 and MediaPipe

This study introduces an innovative real-time AI posture correction service for three major powerlifting exercises: bench press, squat, and deadlift, utilizing YOLOv5 and MediaPipe. Due to the rising popularity of online fitness apps post-pandemic, there is a need for accurate posture correction tools to prevent injuries associated with incorrect exercise forms. The proposed method involves a comprehensive MultiPose Exercise Dataset (MPED), which collects 3D joint coordinate data from multiple angles using a smartphone camera. This data is used to train machine learning and deep learning models for detailed posture classification and real-time feedback. The AI service offers specific corrective feedback for both concentric and eccentric contractions, improving exercise efficiency and safety. The study’s results show that models built with machine learning algorithms generally outperform deep learning models for posture classification, and the proposed detailed feedback system is effective in preventing injuries and enhancing performance. This comprehensive approach ensures that users receive tailored feedback to correct their form in real-time, significantly reducing the risk of injury and promoting better overall fitness outcomes.