摘要
全车运行故障动态图像综合检测系统安装在客整所入段线上,对动力集中型动车组的车底走行部、车侧转向架、车体车窗及车顶部位进行动态检测,自动识别判断关键部件变形、脱落、异物等异常情况。撒砂管是动力集中型动车组的一个重要部件,由于其并不是密封在动车内部,所以容易受到异物的撞击和外力的作用,使得撒砂管高度产生变化,对行车安全和乘客的生命财产安全造成威胁。基于全车运行故障动态图像综合检测系统,提出了一种结合2D图像与3D点云数据处理的撒砂管高度检测方法。首先根据3D点云数据拟合参考平面的平面方程;然后在2D数据上通过深度学习技术定位出撒砂管所在区域的数据并根据2D数据与3D点云数据的对应关系获取到对应的3D点云数据;最后对3D点云数据进行连通域分析和滤波处理提取撒砂管的底面点云数据,并根据提取到的底面点云数据和参考平面方程计算撒砂管高度。实验结果表明,该算法能够克服复杂背景对测量结果的干扰,96%的数据测量误差不超过3 mm,90%的数据测量误差不超过2 mm,所有测试数据中最大测量误差不超过4 mm,满足实际行车过程的检测需求。
The dynamic image comprehensive detection system for the whole train operation fault is installed on the entrance section line of the passenger integration station to dynamically detect the running part of the underbody,side bogie,window and roof of the power centralized EMU,and automatically identify and judge the deformation,falling off,foreign matter and other abnormal conditions of key components.The sand-spreading pipe is an important part of the power-concentrated EMU train.Because it is not sealed inside the EMU,it is vulnerable to the impact of foreign objects and external forces,which makes the height of the sand-spreading pipe change,which is important for driving safety and passengers.The safety of life and property is threatened.Aiming at the problem that the existing research does not use machine vision technology to detect the height of the sanding pipe in the movement process,a non-destructive detection method for the height of the sanding pipe combining 2D image and 3D point cloud data is proposed from the field of machine vision.The method of non-destructive testing of the height of the sand-spreading pipe is to first fit the plane equation of the reference plane according to the 3D point cloud data;then use the deep learning technology to locate the data of the area where the sand-spreading pipe is located on the 2D data,and then according to the 2D data and the 3D point cloud data Corresponding relationship to obtain the corresponding 3D point cloud data,and finally the connected domain analysis and filter processing of the 3D point cloud data to extract the bottom surface point cloud data of the sanding pipe,and calculate the sanding according to the extracted bottom surface point cloud data and the reference plane equation Tube height.Experimental results show that the algorithm can overcome the interference of complex background on the measurement results.96%of the data measurement error does not exceed 3 mm,90%of the data measurement error does not exceed 2 mm,and the maximum measurement er
作者
罗玉胜
谭兆海
赵波
LUO Yusheng;TAN Zhaohai;ZHAO Bo(China Railawy Lanzhou Bureau Group Co.,Ltd.,Lanzhou 730000,China;Chengdu Railway Safety Science & Technology Co.,Ltd.,Chengdu 610091,China)
出处
《技术与市场》
2022年第2期14-18,共5页
Technology and Market
关键词
机器视觉
撒砂管高度测量
目标检测
点云数据处理
machine vision
height measurement of sand-spreading pipe
target detection
point cloud data processing
