Two-stage safety helmet detection in industrial environment using YOLO models

Ensuring that workers in the construction and manufacturing sectors wear helmets is crucial for preventing workplace accidents. This paper presents a two-stage instance detection method for determining helmet compliance using YOLO models. In the first stage, a YOLO with COCO dataset is utilized to detect humans within images. In the second stage, the detected human head positions are classified into "helmet" and "no helmet" categories using a specialized classification model. Our study compares the performance of YOLOv5, YOLOv8, and YOLOv9 models, with results indicating that YOLOv8 achieves the highest Precision and Recall and F1. To further enhance accuracy, a confidence threshold is implemented in the second stage; frames where the model's confidence is insufficient are skipped. This method significantly improves the detection of helmet usage, providing a reliable tool for enhancing safety compliance and reducing the risk of injuries in industrial environments.

Two-stage safety helmet detection in industrial environment using YOLO models Keywords:

component , Helmet detection , YOLO models , Workplace safety , Manufacturing , Deep learning