摘要
安全帽是施工现场的作业人员重要的安全防护工具,且不同身份佩戴安全帽颜色不同。针对作业人员不佩戴安全帽、越界操作等违规行为,本文提出了一种基于改进的Faster RCNN的安全帽佩戴检测及身份识别方法。在原始的Faster RCNN的基础上,会将多个阶段得到的特征层进行融合并进行多尺度检测,同时修改网络本身所应用的候选目标框大小,使得网络模型达到最优。实验表明,佩戴红、黄、白、蓝颜色安全帽,和未佩戴安全帽工人五类目标平均检测准确率达到85.8%,且模型有一定泛化能力和鲁棒性。
Safety helmets are important safety protection tools for workers at the construction site,and helmets of different identities wear different colors.Aiming at violations such as the operator not wearing a helmet and operating outside the boundary,this paper proposes a method for detecting and identifying a helmet based on an improved Faster RCNN.Based on the original Faster RCNN,the feature layers obtained in multiple stages are fused and multi-scale detection is performed.At the same time,the size of the candidate target box applied by the network itself is modified to make the network model optimal.The experiments show that the average detection accuracy of the five types of targets wearing red,yellow,white and blue helmets and workers without helmets reaches 85.8%,and the model has certain generalization ability and robustness.
作者
吴冬梅
王慧
李佳
WU Dong-mei;WANG Hui;LI Jia
出处
《信息技术与信息化》
2020年第1期17-20,共4页
Information Technology and Informatization
