Research On Railway Images Classification And Tread Edge Location Algorithm

With the rapid development of rail transport, the safety of railway operations is getting more attention in that field. Among the track key equipments, the security situation of them, such as rails, fasteners, sleepers, directly determines the safety of the railway operations. With the rapid development of China’s high-speed, heavy haul railway, artificial visual has been difficult to meet the safety requirements of a modern railway operators. The other hand, now the method internationally commonly used is image processing techniques, to achieve the automatic detection of railway track, which is not only high efficient, but also economic, safe, practical. In recent years, by leaps and bounds computer hardware and software technologies, have provided a broader application for automated detection. Therefore, our country must keep up with the international trends, according to local conditions, develop the automated detection system that fits our country’s railway situation.In this paper, it is aimed at one of the image processing algorithms’characteristic that it is highly targeted to carry out research, and the main work done is as follows:Firstly, it makes analysis on the railway images’features, and then based on the characteristics of the light intensity and the light distribution, it divided the railway image into four different types of images, following the order of rudimentary classification and fine classification, then it programs different location algorithm aimed at different kinds of images. The method aims at image processing algorithms’adaptive characteristics effectively, which greatly improves the accuracy of tread edge positioning.Secondly, it proposes comparison algorithm based on the similarity principle. This method cuts out a part of a image having the same resolution as the tread as a template, and then traverses the full image to cut out many parts having the same resolution as the template, to do similarity calculation, generating the similarity curve, based on which it determines the type of the image. The algorithm proved to be strong stability and robustness.Thirdly, through theoretical analysis of histogram equalization, it proves that certain gray values mergered is the main cause of the lack of image detail, and then it judges the mergered gray values, combining with a image’s background part and target part which is divided by Ostu threshold segmentation, and finally deals with neighborhood processing to the picture elements corresponding to the mergered gray values. This method can effectively reduce the lack of detail that cauesd by histogram equalization.Fourthly, by the experimental verification to the sample library, it shows that the railway images classification and tread edge location algorithm advanced in the paper is effective.