Semi-automatic ergonomic posture assessment with deep convolutional neural network

초록

The evaluation of the potential risks of WMSDs (Work-related Musculoskeletal Disorders) in industries have been a challenging as the working environment is complexed, which make it hard to evaluate the posture correctly when occlusions occur. This research presents a new WMSDs (Work-related Musculoskeletal Disorders) evaluation methodology by employing a DLRULA (Deep Learning based Rapid Upper Limb Assessment), aimed at evaluating joint angles robustly in cluttered image, and evaluating the ergonomic assessment automatically. DLRULA has been validated with the ground truth joint angle based RULA grand score using VIA (VGG Image Annotator software) tool, with an observation method based human RULA experts. The experimental result shows a statistical almost perfect match of the Landis and Koch scale (DLRULA-VIA: proportion of agreement = 100%, k = 1, DLRULA-Expert: proportion of agreement = 88.9%, k = 0.87). © IMSCI 2020.All right reserved.

키워드