基于视频的轨道车辆自主定位方法研究  

Vision-Based Absolute Position Extraction Method for Rail Vehicles

作　　者：沈拓[1] 谢远翔 盛峰 谢兰欣 张颖[1] 安雪晖[2] 曾小清[1] SHEN Tuo;XIE Yuanxiang;SHENG Feng;XIE Lanxin;ZHANG Ying;AN Xuehui;ZENG Xiaoqing(Shanghai Key Laboratory of Rail Infrastructure Durability and System Safety,Tongji University,Shanghai 201804,China;School of Civil Engineering,Tsinghua University,Beijing 100084,China;Powerchina Roadbridge Group Co.,Ltd.,Beijing 100070,China;Shanghai Zegao Electrical Engineering Technology Co.,Ltd.,Shanghai 201900,China)

机构地区：[1]同济大学上海市轨道交通结构耐久与系统安全重点实验室,上海201804 [2]清华大学土木水利学院,北京100084 [3]中电建路桥集团有限公司,北京100070 [4]上海泽高电子工程技术股份有限公司,上海201900

出　　处：《同济大学学报（自然科学版）》2024年第2期174-183,共10页Journal of Tongji University:Natural Science

基　　金：国家重点研发计划(2022YFB4300501);上海市科委课题(23DZ2204900);校企战略性合作专项高速铁路绿色智能施工关键技术研究(kh0160020230946、LQKY2022-01-1)。

摘　　要：针对轨道施工车辆自主定位需求,提出一种基于车载前视相机拍摄百米标视频的轨道车辆自主定位方法。该方法首先对YOLOX-s网络进行改进并构建了百米标的目标检测模型,完成对百米标的目标检测;其次,当检测到百米标后,结合图像预处理和卷积循环神经网络(CRNN)网络构建百米标数字文本识别模型,提取百米标的数字文本信息,从而实现对轨道施工车辆的定位。经实验验证该方法能够快速准确定位轨道施工车辆的位置信息。Aiming at the demand of autonomous positioning of rail construction vehicles,this paper proposes a method of absolute position extraction of rail vehicles based on the video of 100-metre markers captured by the on-board forward-looking camera.The method first improves the YOLOX-s network and constructs the target detection model of the 100-metre marker to complete the target detection of the 100-metre marker.Then,when the 100-metre marker is detected,it combines image preprocessing with convolutional recurrent neural network(CRNN)network to construct the 100-metre marker digital text recognition model to extract the digital text information of the 100-metre marker,so as to achieve the absolute position location of rail construction vehicles.The method is verified to be able to quickly and accurately extract the absolute position information of rail construction vehicles.

关 键 词：轨道车辆定位 机器视觉 目标检测 文本识别 

分 类 号：U239.5[交通运输工程—道路与铁道工程] TP391.41[自动化与计算机技术—计算机应用技术]