基于轮廓重构的列车检测图像匹配算法研究

发布时间：2018-06-13 11:00

本文选题：TFDS + 轮廓重构　；参考：《湖北工业大学》2017年硕士论文

【摘要】：近几年来,货运列车运行故障动态检测系统(TFDS)被逐渐推广运用于我国各铁路干线,该系统避免了人工列检的局限性,实现了货运列车的自动检测。然而由于货运列车零件外形多样、外界环境恶劣,导致TFDS系统图像背景复杂、目标区域锁定不稳定。本文以列车关键部位零件闸瓦和缓解阀盖为研究对象,对基于轮廓重构辅助的匹配和基于轮廓重构特征描绘的匹配进行了深入研究。提出一种基于重构轮廓辅助匹配路径规划的动态分级匹配方法。该方法依靠待检图像中具有典型几何特征对象的重构轮廓来规划遍历路径,以轮廓Hu不变矩描绘目标形状特征,并采用高低双阈值门限与灵活动态分级匹配机制相结合的策略。对闸瓦检测图像的实验表明,本文算法在提升遍历效率的同时兼顾了闸瓦区域锁定的准确性,满足了TFDS闸瓦区域快速准确锁定的要求。提出一种基于轮廓有向线段重构与直方图统计的匹配方法。该方法结合基于三角形局部弯曲度评价机制的轮廓有向线段重构,以所有成对有向线段之间有向相对角和归一化当量距离的统计直方图作为目标整体形状特征的描绘,最后通过计算直方图距离衡量目标之间的相似性。以缓解阀盖为对象进行综合实验表明,本文提出的匹配方法对于旋转、平移和缩放等几何变化均具有良好的适应性,实现了目标区域的准确锁定,满足TFDS缓解阀盖检测区域定位的实时性要求。本文结合轮廓Hu不变矩、弧度步长、动态分级匹配、基于三角形单元弯曲度评价机制的动态采样、直方图统计匹配等相关措施,较好的解决货运列车闸瓦和缓解阀盖区域精准快速锁定的问题。文中的闸瓦匹配方法可推广运用于使用闸瓦制动的不同列车,缓解阀盖匹配方法亦可推广运用于具有不规则外形的盘盖类零件检测。
[Abstract]:In recent years, the dynamic fault detection system of freight train operation (TFDS) has been gradually popularized and applied to the main railway trunk lines in China. This system avoids the limitation of the manual inspection and realizes the automatic detection of the freight train. However, because of the various parts of the freight train and the bad external environment, the image background of the TFDS system is complex and the target area lock is locked. In this paper, the key parts of the key parts of the train and the relief valve cover are studied, and the matching based on the auxiliary contour reconstruction and the matching based on the contour reconfiguration feature description is studied. A dynamic hierarchical matching method based on the reconstructed contour auxiliary matching path planning is proposed. The reconstructed contour of typical geometric feature objects is used to plan the traversing path, to describe the feature of the target shape with the Hu invariant moment of outline, and to combine the high and low threshold threshold with the flexible dynamic classification matching mechanism. The experiment on the detection image of the brake shoe shows that the algorithm in this paper is to improve the ergodic efficiency while taking the lock area lock into consideration. The accuracy of the method satisfies the requirement of fast and accurate locking in the TFDS brake shoe area. A matching method based on the contour line segment reconfiguration and histogram statistics is proposed. This method is combined with the line segment reconstruction based on the triangular partial curvature evaluation mechanism, with the opposite angle and the normalized equivalent distance between all pairs of directed lines. The statistical histogram is described as the feature of the overall shape of the target. Finally, the similarity between the targets is measured by the calculation of the histogram distance. A comprehensive experiment with the relief valve cover is carried out to show that the matching method proposed in this paper has good adaptability for the geometric changes such as rotation, translation and scaling, and the accuracy of the target area is realized. It meets the real-time requirement of TFDS relief valve cover detection area positioning. This paper combines the contour Hu moment, the arc step length, the dynamic classification matching, the dynamic sampling based on the triangular element bending degree evaluation mechanism, the histogram statistics matching and other relevant measures, which can better solve the precise and fast locking of the cargo train brake shoe and the relief valve cover area. The method of brake shoe matching in this paper can be used to promote the use of different trains used to brake the brake shoe. The matching method of relief valve cover can also be applied to the detection of plate cover parts with irregular shape.
【学位授予单位】：湖北工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U279.32;TP391.41

【参考文献】