基于图像处理的转辙机接点组打入深度提取算法

Penetrating Depth of Contact Group Extraction Algorithm for Switch Machine Based on Image Processing

转辙机是铁路上改变轨道方向,反映道岔位置的重要器械,转辙机腔内接点组中动接点与静接点的打入深度是判断转辙机是否正常工作的重要指标。由于转辙机内部结构非常复杂,传统工业物件的图像测量方法并不适用,为了提高转辙机检修的准确性、减小人力开销,本文设计了基于图像处理技术,针对广泛使用的ZD(J)9型转辙机的内部打入深度的测量算法。首先该算法通过模板匹配的方式准确定位整个接点组区域,排除其他区域结构的干扰。同时针对拍摄图像可能存在的倾斜问题,提出了基于图像仿射变换的预处理算法将倾斜的接点组图像校正。从图像距离到实际距离需要一个比例尺进行换算,算法在校正后的图像中选取长度明确、畸变较小的物体长度作为比例尺计算的参照物。为进一步精确定位动接点和静接点的位置,设计了基于meanshift的滤波算法,HSV色彩空间中的颜色阈值分割算法,以及按照数字图像连通域的去噪算法。最后,在精确定位的基础上加以一定误差补偿,进行参数计算。本文算法在不同拍摄高度下得到的图像中进行实验,对比了不同拍摄高度下的参数提取效果,对误差原因进行详细分析。

Switch machine is an important instrument to change track direction and reflect switch position in railway. The penetrating depth between dynamic contact group and static contact group is an important index to judge whether switch machine worked normally. Owing to complicated internal structure of switch machine, traditional image measurement methods of industrial objects are not applicable. In order to improve the accuracy of switch machine maintenance and reduce the cost of manpower, this paper designed an algorithm for measuring the internal penetrating depth of ZD(J)9 switch machine based on image processing. First of all, this algorithm can accurately locate the whole contact group region by means of template matching, which eliminated the interference of other region structures. Meanwhile, a preprocessing algorithm based on affine transformation is also proposed to rectify oblique images, after which the object length with clear length and small distortion will be selected as the reference for scale calculation used for converting image distance to actual distance. Some methods help further accurately locate the positions of contact groups including meanshift filtering, color threshold segmentation in HSV color space and a denoising algorithm based on connected domain were also proposed. Finally, with the help of error compensation algorithm, the calculation result can be more precise. This algorithm was tested on images obtained at different shooting heights, revealing the effect of parameter extraction by comparison. The causes of error were also analyzed in detail.