Ergonomic Risk Assessment Using Human Pose Estimation with MediaPipe Pose

This research utilizes Human Pose Estimation (HPE) to detect improper working postures that may lead to Work-Related Musculoskeletal Disorders. These postures are evaluated according to ergonomic principles using the OWAS method, which assesses the back, arms, legs, and the weight of the load during the task with codes indicating risk categories. In the experiment, both Two-dimensional (2D) and three-dimensional (3D) keypoints of body joints obtained from MediaPipe Pose and the angles of the joints relative to 72 different postures, encompassing all combinations of the code for the three body parts, were used as features for setting the criteria in posture prediction. The best classifier obtained from the experiment provided accuracy rates of 90.74% for the back, 92.71% for the arms, and 85.53% for the legs, respectively. When the program was applied to evaluate working postures in a simulated packing station, the accuracy rates for the back, arms, and legs were 70.42%, 82.78%, and 71.25%, respectively. The accuracy in identifying the risk category, which considers all body parts simultaneously, is 77.64%.