Research On Personal Protective Equipment Detection In Tunnel Construction Sites Using Computer Vision And Deep Learning Approaches

Wearing suitable personal protective equipment(PPE)like helmets can avoid most injuries and fatalities on construction sites.Therefore,monitoring PPE is one essential task for construction management.However,the tunnel construction site,which is usually lowlight and full of dust,brings challenges for the intelligent detection of PPE.This study adopted computer vision and deep learning techniques to detect PPE in outdoor and tunnel low-light environments where the main research contents and conclusions are as follows:(1)This study analyzed state-of-the-art object detection deep learning models and their applications on construction sites.(2)For detecting different color helmets and vest,this study trained You Only Look Once(YOLO)family models for obtaining different performance of models.It eased the problem of the limited detection range of existing deep learning models in outdoor environments and improved the performance of the models.The experiment results show that YOLOv5 x has the highest correctness(m AP 86.55%),while YOLOv5 s has the fastest processing speed(52 FPS).(3)To improve the performance of low-light detection,this study tested CLAHE data enhancement algorithms and modified the backbone of YOLOX by adding a Conv Ne Xt network structure and a fourth detection head to enhance the feature learning.Experiments show the improved YOLOX can significantly improve the detection accuracy of small objects,whose m AP achieves 86.49%,4.23% higher than the original model and other deep learning models.(4)This study constructed a high-quality dataset with multiple PPE classes for outdoor construction sites and a novel low-light PPE dataset for tunnel construction.The two datasets provide the basis for training deep learning models.