Safety Helmet Detection at Construction Sites Using YOLOv5 and YOLOR

Abstract

Wearing a helmet is mandatory for workers at construction sites. It is very important for the safety of workers during work. In many scenarios, detecting workers not wearing helmets can prevent possible occupational accidents in time. Recently, with the rapid development of deep learning, convolutional neural networks (CNNs) have been widely applied in many problems including object detection. The constantly evolving object detection technology has resulted in a series of YOLO algorithms with very high accuracy and speed being used in various scene detection tasks. This paper presents a deep learning approach to solve the above problems. We propose a helmet detection method based on two models, namely YOLOv5 and YOLOR, using a dataset of 900 collected images. The two models are compared and analyzed. The experimental results show that the mAP@0.5 of YOLOR reached 87.3%, significantly larger than that of the YOLOv5 model with mAP@0.5 of only 77.6%, proving the effectiveness of helmet detection using the YOLOR model.