基于改进型YOLOV3安全帽检测方法的研究  

Helmet Detection Based on Modified YOLOV3

作　　者：张勇[1] 吴孔平[1] 高凯 杨旭 ZHANG Yong;WU Kong-ping;GAO Kai;YANg Xu(College of Electrical and Information Engineering,Anhui University of Science and Technology,Huainan Auhui 232000,China)

机构地区：[1]安徽理工大学电气与信息工程学院,安徽淮南232000

出　　处：《计算机仿真》2021年第5期413-417,共5页Computer Simulation

摘　　要：由于现场施工人员不佩戴安全帽的情况时有发生,导致产生很多威胁人身安全的问题,为了解决由于不佩戴安全帽所引发的安全问题,实验尝试把目标识别技术应用到建筑安全领域,提出了一种基于目标算法-YOLOV3(约洛)安全帽检测方法。在此基础之上,采用DenseNet(密集卷积网络)方法处理低分辨率的特征层,在保持特征提取充足的同时降低了计算复杂度,从而提高了算法的检测和收敛性能。并使用K-means(k均值)聚类算法对数据集中的目标边框重新进行聚类,通过自制的增强型安全帽数据集进行训练,得到改进的YOLOV3模型,使其更加适应安全帽领域的应用场景,提高准确率和识别速度。实验将改进后的YOLOV3模型与原YOLOV3模型进行对比,结果表明,改进后的YOLOV3模型在对安全帽检测过程中,比原YOLOV3模型准确率更高,达到了96.5%以上。同时检测速度达到了59.6fps,满足安全帽实时检测的要求。Due to the frequent occurrence of construction workers not wearing safety helmets on site, there are many problems threatening personal safety. In order to solve the safety problem caused by not wearing safety helmet, this paper attempts to apply target recognition technology on the field of building safety and proposes a target algorithm-YOLOV3 helmet detection method. On this basis, the DenseNet method was used to process feature layers with low resolution in the YOLO-V3 network. The feature extraction of capacity was maintained while the computational complexity was reduced, thus the detection performance of the algorithm can be improved. At the same time, the K-means method was used to re-group the target borders, and the accuracy and recognition speed of this model were improved. The improved YOLOV3 model is more adaptable to the application of helmets. The improved YOLOV3 model was compareed with the original YOLOV3 model. The results show that the improved YOLOV3 model is better than the original YOLOV3 model on the accuracy, and the speed can reach to 59.6 fps, which meets the requirements of real-time detection of helmets.

关 键 词：安全帽 约洛 密集卷积网络 均值 实时检测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]