The Vision System for Diagnostics of Railway Turnout Elements

Abstract

The railway traffic usually runs on one or two lines that connect with each other through the railway turnout. Railway turnout due to its design and purpose belongs to the subassemblies that are part of the infrastructure of railways on which there is the highest number of accidents of railway vehicles. Currently, in most countries there are high speed railways for which existing solutions have been rebuilt or new solutions for railway turnouts have been designed. During the passage, the main elements causing possible derailment of the train are the switch point and the crossing. During the passage of the wheel on the switch point and on the crossing, forces appear even twice as high as when driving on the track. The approaches to railway turnout diagnostics based on the image processing obtained from cameras placed at the points of the railway turnout will be presented. The diagnostic system will perform a real-time inspection of the condition of the surface of the switch point and crossing and will determine the geometry values of both components and perform a comparison with the data contained in the technical standards. The approach to railway turnout image processing was divided into three stages, including pre-processing, extraction of features and processing of functions obtained, using neural networks, the architecture of which will be presented in the work. The proposed method gives quick and real diagnostic results.