文献类型：期刊文献

中文题名：基于多任务级联卷积神经网络的交通标志检测

英文题名：Traffic sign detection based on multi-task cascaded convolutional neural network

作者：王弘宇[1];张雪芹[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2022

卷号：43

期号：1

起止页码：210

中文期刊名：计算机工程与设计

外文期刊名：Computer Engineering and Design

收录：CSTPCD;;北大核心:【北大核心2020】；

语种：中文

中文关键词：交通标志检测;多任务级联模型;特征提取网络;区域生成网络;mask分支

外文关键词：traffic sign detection;multi-task cascade model;feature extraction network;region proposal network;mask branch

摘要：为解决复杂驾驶环境中小目标交通标志的高精度检测问题,基于Cascade R-CNN级联模型提出一种多任务级联模型GA-CMF R-CNN(guided anchoring-cascade mask flow R-CNN)。采用ResneXt101(32×4d)-FPN作为特征提取网络,确保特征图的语义信息和分辨率信息;采用GA-RPN作为区域生成网络,提高网络特征表达能力;模型融合Mask R-CNN的mask分支,在级联的mask分支间添加信息流,通过对先验框中的目标进行语义分割,提高检测精度。在公开交通标志检测数据集上的测试结果表明,该模型能有效提高复杂环境下小目标交通标志的检测和识别精度。
To solve the problem of high-precision detection of small target traffic signs in a complex driving environment,a multi-task cascade model GA-CMF R-CNN(guided anchoring-cascade mask flow R-CNN)based on the cascade R-CNN was proposed.ResneXt101(32×4d)-FPN was used as the feature extraction network to ensure the semantic information and resolution information of the feature map.GA-RPN was adopted as the region generation network to improve the network feature expression ability.Mask branch from Mask R-CNN was merged and information flow between the cascading mask branches was added to improve the detection accuracy by semantically segmenting the target in the prior anchor.The test results on the public traffic sign detection dataset show that the model can effectively improve the detection and recognition accuracy of small target traffic signs in complex environments.