Automatic Ovako Working-Posture Analysis System (OWAS) Application Using Deep Learning

Abstract

The imperative for industries to adhere to national employee safety standards, particularly concerning work-related musculoskeletal disorders (WMSDs), has necessitated robust working posture assessments. Traditionally, the Ovako Working Posture Analyzing System (OWAS) has been employed for this purpose. However, this conventional method, which relies on direct observation and limited sample sizes, faces challenges in terms of accuracy and scalability. This project aims to address these limitations by introducing a state-of-the-art deep learning model into the assessment process. Utilizing a Human Pose Estimation framework trained on a ResNet-18 architecture, the model leverages transfer learning techniques with the COCO datasets for enhanced generalization capabilities. The methodology integrates seamlessly with OWAS by selecting key-points pertinent to its risk factors. These key-points are algorithmically linked to form vectors that serve as the basis for calculating posture angles. Subsequently, these angles are translated into risk scores, which are then mapped onto OWAS risk levels. The resultant approach promises unprecedented accuracy and scalability in working posture assessment, potentially revolutionizing compliance with occupational safety norms.

Keywords: Ovako Working Posture Analyzing System (OWAS), work-related musculoskeletal disorders (WMSDs), deep learning, Human Pose Estimation, occupational safety norms