Research on Identification and Detection of Unsafe Behaviors of Construction Workers Based on Deep Learning

Main Article Content

Meiyu Zhang
Hongming Chen
Xuefeng Han

Abstract

In order to improve the safety management level of construction sites, prevent and reduce the occurrence of building safety accidents, this article uses deep learning methods to study these unsafe behavior recognition and detection techniques. The most typical hazardous behavior is not wearing a safety helmet. However, on-site personnel often neglect to wear helmets due to various reasons. In this study, the target detection algorithm is applied to monitor helmet-wearing. The YOLOX algorithm is selected as the basic detection model and improved by combining the construction site environment and helmet detection characteristics, meeting the real-time monitoring needs of helmet-wearing. Comparison experiments before and after improvement were conducted on the self-constructed helmet dataset, verifying the performance of the improved YOLOX network model. The results show that the average accuracy of the enhanced network model on the helmet-wearing dataset increased to 89.12%, showing a better detection effect.

Article Details

Section
Special Issue - High-performance Computing Algorithms for Material Sciences