Towards Safer Construction Sites: A Preliminary Study on a Vision-Based Cyber-Physical System for Automated Hazard Detection

The construction industry is inherently associated with various hazards that significantly impact worker safety and well-being. Traditional manual hazard detection methods are often compromised by human errors, fatigue, and oversight. This paper presents a vision-based cyber-physical system (CPS) to enhance hazard identification in construction environments. The proposed CPS comprises a physical layer, integrating a network of cameras, and a cyber layer powered by the Gaussian splatting technique and segment anything model (SAM). The system captures on-site images for 3D reconstruction, uses SAM to detect hazards including unsecured machinery and hazardous materials, and then delivers alerts to workers via their smartwatches. To evaluate its effectiveness, the system was deployed in an indoor construction environment. Results showed that the CPS achieved over 95% hazard detection accuracy and an 87.1% pixel accuracy (PA) in identifying hazards. These findings demonstrate the system's potential to reduce accidents and enhance overall safety in the construction industry.