Development of vision based methods for diagnosis of rail surfaces, turnouts, and level crossings in railway

The railway network is rapidly growing in the world, due to the spread of high-speed trains and the safety of railway transportation. As railway transportation is widespread in the whole world, the importance given to the safety of railway transportation and the maintenance of railway components is also increasing. The most basic measure to be taken to reduce railway accidents is to check at regular intervals of railway components. Diagnosis of rails and track components by contact monitoring techniques can be a great burden in terms of both time and human resources. Controlling the rail surface using touching manual measurement techniques will both damage rail surface and cause railway access to be temporarily disabled. In many methods available in the literature, rail line and rail components are diagnosed using contactless image processing techniques. In this thesis study, three basic methods for image processing based monitoring, diagnosis and detection of the railway track have been developed. In the first method, defects such as abrasion, fracture, carving occurred on the rail surface have been detected by using image processing techniques. In order to carry out the mentioned process, the rail line in the image as the input parameter is determined first. In the second method, railway scissor transitions have been detected using SVM (Support Vector Machine) and some image processing techniques. In the last method, the detection of the level crossings and the foreign objects that can cause accidents on the level crossing are detected by image processing based techniques. As a result, the railway track has monitored, diagnosed and detected with image processing based methods in this thesis study. This study has been supported by a research project in TUBITAK 1001 program under Project No: 114E202 and the studies were concluded with various national and international publications.