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

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.

The results were presented at the Construction Research Congress (CRC) 2024 and published in the proceedings (link to article).