ENHANCING SAFETY AT CONSTRUCTION SITES THROUGH ADVANCED FASTER R-CNN MODELING

Abstract

Accidents and deaths at construction sites are on the rise, including incidents where buildings under construction have unexpectedly collapsed, impacting nearby vehicles and workers. There is an increasing demand for more manpower and authorized personnel for safety inspections at these sites. To mitigate accidents and fatalities, the use of sensors, mobile technologies, and machine learning algorithms is essential. This paper presents a model designed to identify safety hazards based on construction workers' compliance with wearing Personal Protective Equipment (PPE) during working hours. The model has been integrated into a website as an image surveillance tool to monitor safety conditions at construction sites. The model was developed using 7,000 images from the MIT Places Database (for scene recognition) as a training dataset and 400 images sourced from Google for testing. It was built using TensorFlow 1.15, Python scripting, and the Faster R-CNN algorithm. The evaluation results of the model indicate a 75% accuracy rate. The novelty of this model lies in its integration of four components of PPE and its practical application for mobile usage. Supervisors at construction sites can utilize their mobile phones and drones as surveillance tools to enhance the safety of their workers. In future developments, the model will also incorporate external factors contributing to accidents and fatalities, such as environmental conditions, electrical hazards, and falls.