Research on Train Positioning Algorithm with Special Rail Characters

Abstract

Locating exactly where a train is on a track is now a major concern for railway companies. By training the charac-teristic objects from the picture samples along the railway with YOLO v5 to generate the recognition template, the characteristic images containing characteristic objects can be selected as the positioning points. Then Compile the identity code (ID code) of the positioning points' pictures by using the location information of the characteristic objects. Match the detected pictures' ID code with positioning pictures' ID code through similarity, and the recognition can be completed if the similarity is higher than the set threshold. Finally, by fetching the location information of the positioning point, the train can identify it position. Through a series of methods such as changing the shooting angle, sharpness and contrast of the positioning point images, the testing set is expanded, and the YOLO v5 based positioning algorithm can be measured its optimal model. The experimental results show that when the similarity threshold is 0.58 and the confidence limit is 0.6, the train positioning model has the best performance, and the success rate of positioning is 97.6 %.