Automated evaluation of unsafe working postures in lifting and carrying heavy objects in construction using a CNN deep learning model

Abstract

Lifting and carrying heavy objects takes place very commonly and with high intensity on construction sites. Wrong working posture likely leads to musculoskeletal disorders, then risks of health damage and illness for workers. This study seeks to propose a method to automatically assess unsafe postures of lifting and carrying heavy objects by combining the RTMPose deep learning model to detect people from videos and a convolutional neural network (CNN) model to automatically extract, evaluate and classify the worker’s posture skeleton frames into two states “safe posture” and “unsafe posture”. The study also provided two datasets of the worker’s skeleton posture skeleton frames of lifting and carrying heavy objects for further research and applications. The proposed method has been experimentally tested with good results. Finally, some ways to apply this method in
managing and controlling occupational health risks in practice have also been discussed.