Hybrid algorithm for the detection and recognition of railway signs

Abstract

This paper presents an application of a hybrid algorithm for detection and recognition of railway signalling. The proposed algorithm has been implemented in a safety application which supports train drivers work. The algorithm is dedicated to performing the tasks of early warning about events related to signalling at railway infrastructure. Simulation tests have been performed in order to find an optimal solution of the problem with the use of classic digital image processing algorithms, the Haar cascade classifier and neural networks. As a result of the conducted research, a hybrid algorithm is developed that uses the You Only Look Once (YOLO) neural network to detect the position of semaphores in the image and the classic methods of digital image processing to recognize the colour light signals. As a part of the research, a database of test images was prepared along with test applications. Research methodology is based on Lane Startup method.