【摘要】 在工地生产与建设中,工人由于未正确规范佩戴安全帽而发生的安全事故时有发生,为了保护工人的安全与降低由于为佩戴安全帽就进入施工现场而引起的意外事件的发生,提出了一种基于改进YOLOv3算法的安全帽佩戴检测方法。通过反卷积模块对网络进行上采样而增加网络在小物体上的表达能力来提高网络的检测准确度。理论分析与实验结果表明,改进的D-YOLOv3网络在检测速率不下降的情况下,mAP(Mean Average Precision)达到了88.8%,其准确率相较于YOLOv3有一定的提高,所提出的算法满足安全帽检测任务的准确性与实时性的要求。更多还原

【Abstract】 In the production and construction of the construction site, the safety accidents that occur due to workers not wearing the safety helmets have occurred from time to time. In order to protect the safety of the workers and reduce the occurrence of accidents caused by entering the construction site for wearing the safety helmets, we have proposed A detection method of helmet wearing based on improved YOLOv3 algorithm. Through the up-sampling of the network through the deconvolution module to increase the network’s ability to express on small objects to improve the detection accuracy of the network. Theoretical analysis and experimental results show that the improved D-YOLOv3 network achieves 88.8% mAP(Mean Average Precision)without a decrease in the detection rate, and its accuracy is improved compared to YOLOv3. The accuracy and real-time requirements of hard hat inspection tasks.更多还原