Integration of Computer Vision and Convolutional Neural Networks in the System for Detection of Rail Track and Signals on the Railway

Abstract

One of the most challenging technical implementations of today is self-driving vehicles. An important segment of self-driving is the ability of the computer to “see/detect” objects of interest at a distance which enables safe vehicle operation. An algorithm for the detection of railway infrastructure objects, namely, track and signals, is proposed in this paper to enable detection of signals which are relevant for the track the train is moving along. The algorithm integrates traditional computer vision (CV) algorithms, including Canny edge detection, Hough transform, and You Only Look Once (YOLO) algorithm, based on convolutional neural networks (CNNs). Each of the concepts (CV and CNNs) deals with a different object of detection which together form a unique system that aims to detect both the rails and the relevant signals. This approach ensures that the artificial intelligence (AI) system is “aware” of which route the signal belongs to. The reliability of the proposed algorithm in detection of a relevant signal, verified by the performed tests, is up to 99.7%. The metric method used for validation was intersection over union (IoU). The obtained value of IoU applied on the entire validation dataset exceeds 0.7. Calculated values of average precision and recall were 0.89 and 0.76, respectively. The algorithm created in this way solves the problem of detection of relevant signals along the train route, especially in multitrack scenarios such as stations and yards.

A Hybrid Retina Net Classifier for Thermal Imaging

“…Classification of an object is a major problem in computer vision [27]. Design is tasked by localizing objects in a thermal image and at the same time dividing them into various groups.…”

Aerial Footage Analysis Using Computer Vision for Efficient Detection of Points of Interest Near Railway Tracks

“…An algorithm for the detection of railway infrastructure objects, namely, track and signals, is proposed by [15] to detect signals relevant to the track and check if the train is moving along. The convolutional neural network based on the You Only Look Once technique, Hough transform, and other well-known computer vision algorithms are included in the algorithm.…”

“…According to the experiments that were run, the proposed algorithm has an up to 99.7% accuracy rate for signal detection. Although this paper uses an algorithm similar to that used in [15], the target POI are different, and the application requires a more lightweight model to perform high-speed detection, which would not be possible with the same model as in [15].…”

Onboard Sensor Systems for Automatic Train Operation

“…The annotation of signals does not include the association with the belonging tracks. Since exceptions to the rules for signal placement exist, these associations can not be easily derived [19]. Furthermore, OSDaR23 does not provide the classification of signal states.…”

“…Although the detection of signals is ensured by automatic train stop in case of GoA1, they still have to be recognized from the vehicle. The challenge of signal detection also includes detection of tracks and their assignment to the signals [18], [23]. From GoA2 on, signals do not need to be detected and are transmitted by cab signalling when used with ETCS.…”