Detect Sitting Pose for Ergonomics Assessment with Image Processing

Abstract

This paper presents a method for detecting and evaluating preliminary sitting postures based on RULA and ROSA theories using anthropometric data captured through a user's camera. The method utilizes Caffemodel to identify six key reference points head, shoulders, elbows, hips, knees, and feet and determines body axis directions. Users' sitting postures are analyzed by calculating angles between these reference points, and specific desk equipment angles are also assessed. In experiments with 68 participants, results were compared to expert evaluations, revealing an average error of 37.70 across all reference points, with blue-dressing detection achieving the highest accuracy. Among the four angles analyzed, arm angle detection showed the lowest percentage error at 44.12 %. The findings demonstrate that detecting reference points and analyzing angles can provide effective data, supporting the development of more advanced programs for ergonomic assessment and improving the evaluation of sitting postures in diverse environments.