基于局部特征描述的高速铁路接触网斜撑套筒定位与故障检测

摘要

针对高速铁路接触网支撑装置斜撑套筒部件的螺钉松脱与脱落问题,提出一种基于采用局部特征描述的统计模式识别算法和螺钉灰度分布规律的图像检测方法.首先计算斜撑套筒部件正负样本的HOG特征,训练基于AdaBoost算法的级联分类器,实现接触网支撑装置图像中斜撑套筒部件的定位识别.为减少误匹配率,采用支持向量机分类器与AdaBoost分类器级联的方法完成目标的精确定位.对定位得到的斜撑套筒子图像利用Hough变换和边缘检测寻找分割直线分离螺钉和套筒,使螺钉可以被单独分析.实验表明,本文方法能够较准确地实现斜撑套筒2种不良状态的检测.%Aiming at detecting the loosening and falling off of screws on the diagonal tubes of the support devices of the Overhead Contact System (OCS) of high-speed railway,an image detection method was proposed based on the statistical pattern recognition algorithm based on local feature description and the grayscale distribution characteristics of the screw.Based on the calculation of the HOG characteristics of the positive and negative samples of the diagonal tubes,a cascade classifier was trained using AdaBoost algorithm to localize and recognize the diagonal tube components in the image of OCS support device.In order to reduce the false matching rate of detection,an SVM classifier cascaded behind AdaBoost classifier was trained to achieve the accurate location of the target.For the located diagonal tube sub-image,the Hough transform and edge detection were adopted to find the dividing line to separate the screws from the tubes so that these screws can be analyzed separately.The experimental results showed that the proposed method can detect two defective conditions of diagonal tubes accurately.