用于动车组故障检测的图像识别算法

摘要

The aim of TEDS was to detect the trouble of moving EMU images based on the matching between present image and previous one. However, the traditional TEDS was easily affected by some related issues such as speed, light and so on. This article proposed an algorithm to recognize the trouble of EMU image by adaptive fusion local and global matching. First, the image was aligned according to the carriage. And then the SIFT matching was adopted to accomplish the rough detection. Based on such detection, the template matching was further used to locate the trouble of EMU image. To reach the requirement of multilevel trouble alarm, we ifnally deifned the trouble level according to the different components. The experimental result showed that the proposed algorithm could detect the trouble of moving EMU effectively.%动车组运行故障动态图像检测系统（TEDS）通过比对现场图像与其历史图像，实现列车运行状态的实时监测和自动报警。由于不同时间采集的图像存在一定程度的差异性，使得单纯基于SIFT特征匹配的故障识别算法误报率较高。为此，本文提出一种自适应融合局部和全局匹配的图像故障识别算法：将图像以车厢为基准对齐配准；基于SIFT特征匹配，通过局部比对粗略定位故障区域；以上述区域为模板，搜寻历史图像以精准定位故障位置。实验结果表明，本算法能有效地分析和预警运行动车组的异常情况，使得系统用户可及时发现重大故障，提升动车运营质量。