Algorithm for the extraction of selected rail track ballast degradation using machine vision

A number of physical methods are used to survey railway track ballast to assess its degradation as a function of deposition. Simulation tests on track models are also conducted. These testing methods, which are generally labour-intensive and expensive, provide an accurate understanding of the extent of ballast degradation. However, the impact of inadequate maintenance can be observed, even on the surface. Therefore, it seems natural in this case to use image registration. State-of-the-art machine vision systems of track geometry cars provide the means to do this. Obtained ballast images provide a baseline for evaluating its level in relation to sleepers. However, no information is available on other signs of track degradation, such as overgrown vegetation (weeds) or the so-called local muddy areas, which are generally a consequence of poor drainage and a lack of subgrade insulation. These degradations are observed to generate distinctive colour images that are superimposed on the overall image of the ballast surface. They differ in colour and shape. Hence, the authors used this phenomenon to develop an algorithm for the extraction of ballast degradation images based on RGB imaging. Surface descriptors have also been offered to assess these degradations. Extensive measurement material from the railway lines was used to conduct survey experiments based on the examples. The results clearly demonstrate the high success rate of the applied method.