Deteksi dan Evaluasi Gerakan Angkat Beban Menggunakan Computer Vision dengan MediaPipe Pose Estimation

Abstract

Skeletal muscle toning exercises are mandatory for all individuals. One of the effective exercises is compound exercises; these exercises involve more than one joint and more than one muscle. One of the movements is a squat. A squat is very crucial for bone and muscle health for the prevention of sarcopenia because the movement is efficient to train many muscles at once but has a high risk of injury if done with improper posture. The movement evaluation process generally requires direct supervision from a professional trainer, which is costly and time-consuming. This study aims to develop a real-time squat movement evaluation system using computer vision technology through the MediaPipe framework. In contrast to artificial intelligence classification-based detection approaches, this system uses a rule-based method that relies on geometric calculations of angles between joints to validate movements without requiring a training process and external datasets that will make the system lighter. The system works by extracting 33 body landmark points from video captures, then processing the coordinates of the shoulders, hips, knees, and ankles using trigonometric formulas to calculate the angles. The evaluation focused on monitoring hip depth and back posture alignment. Test results indicated that the applied rule-based logic was able to detect postural errors and automatically calculate repetitions with a very small average difference in angular accuracy compared to manual measurements.