基于改进Cascade R-CNN的交通标志牌识别  

Traffic signs recognition based on improved Cascade R-CNN

作　　者：徐国整 周越[2] 董斌 廖晨聪 XU Guozheng;ZHOU Yue;DONG Bin;LIAO Chencong(School of Naval Architecture,Ocean and Civil Engineering,Shanghai Jiaotong University,Shanghai 200240,China;School of Electronic,Information and Electrical Engineering,Shanghai Jiaotong University,Shanghai 200240,China;School of Civil Engineering,Southeast University,Nanjing 211189,China)

机构地区：[1]上海交通大学船舶海洋与建筑工程学院,上海200240 [2]上海交通大学电子信息与电气工程学院,上海200240 [3]东南大学土木工程学院,江苏南京211189

出　　处：《传感器与微系统》2021年第5期142-145,153,共5页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(51678360)。

摘　　要：针对雨、雪、雾天等恶劣环境下,交通标志容易被遮挡,且目标较小,难以被高精度识别以及定位的问题,提出先粗检测再精确检测的策略,并采用改进的级联(Cascade)R-CNN:优化锚设计、在线难例挖掘和多尺度训练,同时用图像去雾和增亮算法进行数据增强,最后选用2个不同的骨干网络的模型进行融合。结果表明:在基于虚拟仿真环境下的自动驾驶交通标志识别大赛提供的数据集上,提出的算法表现出优异的泛化能力和准确率,并在指标F1分数达到了0.9972,有效地克服虚拟场景中不同的天气状况和行人状况等干扰因素,实现了道路周边交通标志牌的精确识别。Aiming at the problem that in harsh environments such as rain,snow,fog,etc,traffic signs are easily blocked,and the target is small,which is difficult to be accurately identified and accurately positioned,a strategy of coarse to fine(C to F)is proposed and an improvs cascade region with convolutional neural network features(R-CNN)network is adapted,which includes improved anchor design,online difficult case mining and multi-scale training.At the same time,the image enhancement is performed by image brightness enhancement and defogging algorithm.Finally,two different backbone network models are used for fusion.The results show that the proposed algorithm performs excellent generalization ability and accuracy in the dataset provided by the automatic driving traffic sign recognition competition based on virtual simulation environment and scored 0.9972 on the index F1.It can effectively overcome the interference factors such as different weather conditions and pedestrian conditions in virtual scenes and accurately identify the traffic signs around the road.

关 键 词：智能交通 交通标志牌识别 级联(Cascade)R-CNN 

分 类 号：U495[交通运输工程—交通运输规划与管理] TP391[交通运输工程—道路与铁道工程]