Preprocessing Algorithms of Vision Research Damage to Concrete Railway Sleepers

Abstract

Th e paper presents algorithms for the preprocessing of the visual method of detecting damages of concrete railway sleepers. It starts with acquiring images of the surface of the sleepers, by selecting the recorded frames of the images. Th en, the color image is transformed into monochrome, so as to obtain the highest contrast possible. Th e Kanan and Cottrell algorithms were used for this purpose. A simple way is to segment the damage images of the sleepers, by thresholding, in order to binarize them. However, more elaborate algorithms are recommended. For this purpose, images are denoted using a median fi lter and further morphological operations to extract the edge of damage. For this purpose, noise is removed from images using a median fi lter, and morphological operations are carried out, to extract the edge of damage. In addition, texture images of the surface of the sleepers are used, removing them from the visual content. As the criterion for selecting the preprocessing algorithm, the shape of the image histogram and its standard deviation were assumed. Such prepared images form the basis for further assessment of the size of damages (cracks and voids) and classifi cation of concrete sleepers.