Parseh Designers Transportation Research Institute Quarterly Journal of Transportation Engineering 2008-6598 17 1 2025 09 23 Identifying the Path of the Train to Detect the Obstacle in the Railway using Deep Learning Identifying the Path of the Train to Detect the Obstacle in the Railway using Deep Learning 5093 5117 221064 10.22119/jte.2025.480785.2718 FA Fatemeh Khazaee M.Sc., Department of Control Engineering, School of Electrical Engineering, Iran University of Science and Technology, Tehran, Iran Mohammad Reza Mohammadi Assistant Professor, Department of artificial intelligence, School of Computer Engineering, Iran University of Science and Technology, Tehran, Iran Hossein Blandi Professor, Department of Control Engineering, School of Electrical Engineering, Iran University of Science and Technology, Tehran, Iran Journal Article 2024 09 29 <span style="font-size: 11.0pt; line-height: 115%; font-family: 'Times New Roman',serif; mso-fareast-font-family: Calibri; mso-ansi-language: EN-US; mso-fareast-language: EN-US; mso-bidi-language: FA;">In this paper, a method for obstacle detection in railways is presented. This process can be divided into three stages: In the first step, all the objects in the image are identified by the YOLO network. In the second step, it is necessary to identify all the rails in the image, for this purpose, the UNet network is trained on the Railsem19 dataset. In the continuation of this step, the train path must be separated from the identified rails; Therefore, the design and simulation of a kind of train path identification algorithm has been done, which in addition to the ability to identify the train path in conditions where the rails are completely separated, in some special conditions such as the areas in front of the Railroad switch where the rail is divided several branches, also has the ability It detects the direction of the train. In the third step, the correspondence between the train trajectory and the location of objects obtained by YOLO is checked. If there is a subscription, the obstacle is in the path of the train and otherwise, the path is considered safe. The proposed obstacle detection algorithm, after simulation, was evaluated on a dataset containing 1664 images and was able to detect obstacles with an accuracy of about 87%. Then, the results of the proposed Railroad switch detection algorithm were analyzed and it was observed that this algorithm also has the ability to identify the path of the train with high accuracy (accuracy close to 99%).</span> <span style="font-size: 11.0pt; line-height: 115%; font-family: 'Times New Roman',serif; mso-fareast-font-family: Calibri; mso-ansi-language: EN-US; mso-fareast-language: EN-US; mso-bidi-language: FA;">In this paper, a method for obstacle detection in railways is presented. This process can be divided into three stages: In the first step, all the objects in the image are identified by the YOLO network. In the second step, it is necessary to identify all the rails in the image, for this purpose, the UNet network is trained on the Railsem19 dataset. In the continuation of this step, the train path must be separated from the identified rails; Therefore, the design and simulation of a kind of train path identification algorithm has been done, which in addition to the ability to identify the train path in conditions where the rails are completely separated, in some special conditions such as the areas in front of the Railroad switch where the rail is divided several branches, also has the ability It detects the direction of the train. In the third step, the correspondence between the train trajectory and the location of objects obtained by YOLO is checked. If there is a subscription, the obstacle is in the path of the train and otherwise, the path is considered safe. The proposed obstacle detection algorithm, after simulation, was evaluated on a dataset containing 1664 images and was able to detect obstacles with an accuracy of about 87%. Then, the results of the proposed Railroad switch detection algorithm were analyzed and it was observed that this algorithm also has the ability to identify the path of the train with high accuracy (accuracy close to 99%).</span> https://jte.sinaweb.net/article_221064_209af063c6e073cff3385b485b2e02e1.pdf