A review of machine learning techniques for ergonomic risk assessment based on human pose estimation

Human pose estimation (HPE) has emerged as a vital tool for automating ergonomic risk assessment (ERA), enabling more effective evaluation of employees’ occupational health and safety. Observation-based ERA techniques are becoming more effective at identifying and reducing musculoskeletal injuries associated with the workplace by utilizing computer vision and machine learning technologies. The study thoroughly investigates recent developments in data collection methods, including depth cameras, marker-based motion capture, and deep learning-based pose estimation techniques. Along with methods for connected body poses, 2D and 3D HPE, and ergonomic risk categorisation, it examines developments in depth cameras, marker-based motion capture, and deep learning-based pose estimation. The efficiency of many ERA methods in determining posture-related risks is assessed. For this review study, we selected relevant studies using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Of the 210 research articles collected from the IEEE Xplore, Web of Science, and Scopus databases, 32 satisfied the review requirements for in-depth examination. From these articles, we observed that the deep learning-based HPE systems produced promising accuracies, but they struggled during real-time processing and occluded images. We investigate the potential of techniques that can strike a balance between performance and speed. The effects of HPE on ergonomics are examined, with an emphasis on how it might enable automated risk assessment systems, increase worker safety, and enhance productivity. The survey concludes by exploring future research possibilities, including the integration of multi-modal sensing, domain adaptation for various industries, and the creation of real-time, artificial intelligence-driven ergonomic monitoring systems.