面向监控视频的受电弓与接触网支柱检测

发布时间：2018-03-25 12:22

本文选题：受电弓　切入点：接触网　出处：《西南交通大学》2017年硕士论文

【摘要】：当前我国高速铁路事业正在快速发展,"四横四纵"网络已基本形成,运行车次和速度都在不断增加,铁路的安全运行也越来越受到重视,而供电系统的安全在这中间扮演着关键角色。为了满足不断提高的对铁路供电系统安全检测和监测的要求,缓解人工检测压力,实现自动化、智能化的弓网系统安全巡检,基于图像处理技术的检测和监测手段越来越得到关注。本文的研究工作是按照6C系统中的接触网安全巡检装置和受电弓滑板监测装置的技术规范来展开的。本文算法以动车组车顶图像和接触网巡检图像为实验数据,利用图像处理和机器学习的方法实现了对图像中的目标设备的智能检测提取,最后通过实验测试也验证了本文所提出的算法的有效性。本文的主要工作及创新内容包括以下几个方面:在对图像的预处理过程中,首先研究采用受限对比度自适应直方图均衡化算法(Contrast Limited Adaptive Histogram Equalization,CLAHE)对存在雾气影响、对比度不明显的图像进行图像增强处理。然后,结合Hough变换和Canny算法对车顶图像进行倾角检测,再用透视变换进行图像矫正。最后,利用旋转投影法对接触网巡检图像进行倾角检测,再用仿射变换进行接触网图像矫正。在受电弓检测中,本文采用Sobel算子和形态学操作对受电弓区域进行粗提取。然后,利用Paralleled-Gabor变换提取受电弓区域的方向性特征。最后研究利用多个支持向量机(Support Vector Machine,SVM)分类器的决策融合方法实现受电弓区域的精确检测提取。在接触网支柱检测中,研究了采用检测图像灭点的方式得到接触网图像的透视信息。然后,根据铁轨与支柱的相对位置关系,利用透视信息得到支柱区域位置并采样得到支柱疑似区域图像。最后采用卷积神经网络实现对巡检图像中接触网支柱区域的检测提取。本文对现有动车组车顶图像和接触网巡检图像数据集进行了实验测试。结果表明,本文算法具有较好的适用性,得到了理想的识别率,验证了本文算法具有一定的工程应用价值。
[Abstract]:At present, high speed railway is developing rapidly in our country. The network of "four horizontal and four vertical" has been basically formed, the number of trains and the speed are increasing, and the safe operation of railway has been paid more and more attention. The safety of the power supply system plays a key role in this process. In order to meet the increasing requirements for the safety detection and monitoring of the railway power supply system, relieve the pressure of manual inspection, and realize automatic and intelligent safety inspection of the pantograph and catenary system, More and more attention has been paid to the detection and monitoring methods based on image processing technology. The research work in this paper is carried out according to the technical specifications of catenary safety inspection device and pantograph slide monitoring device in 6C system. The algorithm takes the roof image of the EMU and the patrol image of the catenary as the experimental data. The method of image processing and machine learning is used to realize the intelligent detection and extraction of the target equipment in the image. Finally, the effectiveness of the proposed algorithm is verified by experimental tests. The main work and innovations of this paper include the following aspects: in the process of image preprocessing, In this paper, the constrained contrast adaptive histogram equalization algorithm (Contrast Limited Adaptive Histogram equalization) is first studied to enhance the image with the influence of fog and the contrast is not obvious. Then, the inclination angle of the roof image is detected by combining the Hough transform and Canny algorithm. Finally, using the rotation projection method to detect the obliquity of the patrol image of the catenary, and then the affine transformation to correct the image of the catenary. In the pantograph detection, In this paper, Sobel operator and morphological operation are used to extract the pantograph region. Then, The directional feature of pantograph region is extracted by Paralleled-Gabor transform. Finally, a decision fusion method based on support vector machine (SVM) support Vector machine (SVM) classifier is proposed to detect and extract pantograph region accurately. The perspective information of the catenary image is obtained by detecting the vanishing point of the image. Then, according to the relative position relationship between the rail and the pillar, Using the perspective information to get the position of the pillar area and sampling the image of the suspected pillar area. Finally, using convolution neural network to realize the detection and extraction of the OCS pillar area in the patrol image. In this paper, the existing EMU roof map is presented. The image and catenary patrol image data sets are tested experimentally. The results show that, The algorithm in this paper has good applicability, and the ideal recognition rate is obtained, which verifies that this algorithm has certain engineering application value.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】