AI-enhanced Ergonomics: Revolutionizing Industrial Safety through real-time Posture analysis and PPE detection

Abstract

Despite the continuous advancements in technology and safety regulations, professional accidents in the industry remain a persistent challenge. In the Industry 5.0 era, Artificial Intelligence and cutting-edge Computer Vision techniques are expected to have a transformative impact on industrial environments. In this context, Deep Learning applications can exhibit significant potential in both the primary detection of safety issues and the quick reaction in accidental scenarios. The system proposed in this publication uses tracking algorithms to identify human safety vulnerabilities and early detect people falling or requesting help. Specifically, the first component employs a Transfer Learning technique with YOLOv7 to efficiently determine and detect whether human Personal Protective Equipment is worn correctly. Additionally, the system utilizes YOLOv7 key points detection model to assess the human posture in real time, allowing machines to detect people falling or requesting help. The work concludes presenting experiments that scrutinize the algorithm's detection performance, under varied positions, evaluating the impact of GPUs and cameras and operator’s distance to camera in a pilot aerospace experimental facility.

Keywords: Ergonomics, Deep Learning, Computer Vision, Aerospace, Safety Management

DOI: 10.54941/ahfe1005299