Machine Vision Technology for Locomotives to Identify Railway Colour-Light Signals

Abstract

Improving quality of transport and logistics services in modern conditions is associated with introduction of new technology and improvement of existing technologies of informatization and digitalization of transport. A particular task of introducing digital technologies into the technological processes of railway transport is to increase safety of train traffic. The analysis of the works of domestic and foreign authors on issues of improving safety of train traffic revealed that at present there is a task of introduction of digital devices for analyzing infrastructure objects along the route of a locomotive. This is of importance when increasing speed of trains or when the trips are long, and it is difficult for a person (a driver) to correctly assess the situation and make a right decision. The objective of this work is to develop a method for automatic monitoring of railway infrastructure facilities, by equipping the locomotive with machine vision technology, namely, to ensure the ability to visually control the indications of railway colour-light signals along the route. The locomotive is equipped with a video module for fixing the streaming image along its movement, and with the microprocessor equipment for analyzing the resulting image. As an algorithm for recognizing railway signals in a fixed image, a mathematical apparatus based on models of convolutional neural networks is used. The work performed showed good results in identifying colour-light signals in the analyzed images. Equipping traction rolling stock with technical vision will allow timely identification of track signals, this is especially important on railway tracks where there is no coding in the track circuit, which helps to increase the level of train safety. The development of the presented technology contributes to digitalization of railway transport, which makes it competitive in the world market.