A Computer Vision Approach to Assessing Work-Related Musculoskeletal Disorder (WMSD) Risk in Construction Workers

Work-related musculoskeletal disorders (WMSDs) are a group of painful disorders of muscles, tendons, and nerves caused due to improper work postures prevalent in construction workers. These disorders can cause temporary and/or permanent disabilities and seriously affect workers' livelihoods. Previous research has applied machine learning (ML) for the recognition of WMSD risk. However, previous research used inertial sensors strapped to the body to measure the angle of the body parts. These sensors are expensive and uncomfortable to wear while working. This research aims to eliminate the need for additional hardware through the use of computer vision. In this project, an ML pipeline was built to identify WMSD risk from workers' images. A pre-trained ML framework called Mediapipe Pose was used to generate features from the images. The relative positions of these landmarks were then used as the input for an artificial neural network (ANN) to classify ergonomic and non-ergonomic postures using the supervised learning approach. After hyper-parameter tuning, 100% training and 99.96% validation accuracy were achieved. Finally, the trained model was tested on real-life videos of construction workers and found to perform satisfactorily.