3D Human Pose Estimation Using Blazepose and Direct Linear Transform (DLT) for Joint Angle Measurement

Human pose estimation is a field of computer vision science that studies the determination of joint points in the human body based on images or videos. One of the applications of human pose estimation is to evaluate human movement and performance. In this study, 3D markerless human pose estimation was carried out using the direct linear transform and deep learning Blazepose methods. System testing was carried out on the push-up movement by comparing the data results of the markerless system with the marker-based motion capture system. Push-ups are excellent exercises for developing upper body strength or endurance, such as the arms and shoulders. Push-ups are widely used in rehabilitation or recovery after surgical procedures. The American College of Sports Medicine (ACSM) has established a standard for assessing a person's physical endurance based on the number of successful push-ups. Quantitatively, of all the mean absolute errors calculated, 70.9% were below 30 mm, and for measuring joint angles (elbows, hips, and knees) 43% were below 5 degrees. An error value below 30 mm indicates that the system can be used for human movement analysis.