基于MobileNet-SSD的铁路信号灯检测识别  

Research on Railway Signal Lamp Detection Based on MobileNet-SSD

作　　者：王焕民[1,2,3] 张建柏 裴华艳[5] 蒋兆远 WANG Huan-min;ZHANG Jian-bai;PEI Hua-yan;JIANG Zhao-yuan(Mechatronics T&R Institute,Lanzhou Jiaotong University,Lanzhou 730070,China;Informatization EngineeringTechnology Research Center for Logistics&Transport Equipment of Gansu Province,Lanzhou 730070,China;Industry Technology Center for Logistics&Transport Equipment of Gansu Province,Lanzhou 730070,China;China Railway Lanzhou Railway Bureau Group Co.,Ltd.,Lanzhou 730000,China;School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学机电技术研究所,兰州730070 [2]甘肃省物流及运输装备信息化工程技术研究中心,兰州730070 [3]甘肃省物流与运输装备行业技术中心,兰州730070 [4]中国铁路兰州铁路局集团有限公司,兰州730000 [5]兰州交通大学电子与信息工程学院,兰州730070

出　　处：《兰州交通大学学报》2020年第4期66-70,共5页Journal of Lanzhou Jiaotong University

基　　金：甘肃省高等学校科研项目(2018C-10);甘肃省青年基金(1606RJYA222)。

摘　　要：为解决铁路轨旁信号灯的定位与实时检测问题,在深入分析传统SSD算法与MobileNet模型的基础上,将MobileNet模型的最后平均池化层、全连接层转换为SSD算法的多尺度特征映射层,提出了一种基于MobileNet-SSD的铁路信号灯检测算法.实验结果表明:该算法克服了传统SSD算法对小目标识别不准确、检测实时性较差的问题,检测速度更快,准确率更高;在50 m监控范围内,算法的平均检测准确率达到85%以上,同时具有25帧/s的实时识别能力.To address the issue of location and real-time detection of railway trackside signal lights,on the basis of deeply analyzing the traditional SSD(single shot multibox detector)algorithm and the MobileNet model,a MobileNet-SSD based railway signal light detection algorithm by converting the final average pooling layer and the fully connected layer of the MobileNet model into multi-scale feature mapping layer of SSD is proposed.Simulation results show that the proposed algorithm can overcome the defaults of inaccurate identification of small targets and poor real-time detection of the traditional SSD algorithm.However,the new proposed algorithm have faster detection speed and higher accuracy rate.Within a 50 meters range of monitoring,the average detection accuracy rate of the proposed algorithm is over 85%.Meanwhile,it has a real-time detection ability of 25 frames per second.

关 键 词：铁路信号灯检测 MobileNet-SSD 目标检测 深度学习 

分 类 号：U284.1[交通运输工程—交通信息工程及控制]