Abstract :

Real-time worker helmet detection using deep learning is an automated safety measure designed to reduce workplace accidents. This research implements the YOLO (You Only Look Once) object detection model to identify workers and verify helmet usage in live video streams. By deploying YOLOv5 with optimized training on custom helmet and no-helmet datasets, the system achieves high accuracy and low latency. The framework supports edge devices enabling on-site monitoring without heavy computation. Experimental results demonstrate detection rates exceeding traditional methods. Precision and recall metrics show strong performance in diverse lighting and occlusions. Integrating this system with alert modules enhances safety compliance. Overall, the approach provides a scalable, robust solution for construction site safety.