Real- Time Worker’s Safety Monitoring On Scaffolding Using Vision Based System

Abstract

The construction industry presents its problems regarding safety maintenance, particularly in matters of worker falls. In this case, falls are projected to grow from 33-38%. This project intends to reduce these dangers by developing a vision-based real-time worker’s safety monitoring system that ensures proper use of Personal Fall Arrest Systems (PFAS) on scaffolding. PFAS patients must wear helmets, harnesses, and lanyards because their proper placement is critical in avoiding injury due to falls. The methodology starts with collecting an image dataset. Every image is painstakingly labelled to show all instances of the specific PFAS components and their relevant states. Normalization is performed to match the standards of image inputs with the selected object detection algorithm, YOLOv11. The dataset is divided into training, validation, and testing units, which is representative of the actual situation. This guarantees effective training and testing of the YOLOv11 model for reliable and accurate real-time detection. The verification tests are conducted using the F1 score metric, which evaluates the model in a holistic way. After configuration, the trained YOLOv11 model is deployed on Open CV in the Visual Studio Code (VS Code) environment to enable real-time detection. The ultimate objective is to deliver a system that significantly reduces fall accidents, enhances worker safety, and facilitates compliance with stringent construction site safety standards.