Autonomous Train Operation Obstacle Detection Based on Night Vision System

Abstract

This paper presents the development of an obstacle detection system for autonomous train operation (ATO) using a night vision system. The focus is on the development of the ATO for freight transport that operates in low light and night conditions. An experimental setup featuring an Intensified Charge-Coupled Device (ICCD) camera was employed to acquire images under low-light conditions. A novel computer vision algorithm was developed to ensure robust and reliable obstacle detection. The methodology begins with the detection of rail tracks, which are subsequently used to define a Region of Interest (ROI). Within the ROI, the rail tracks are analyzed for interruptions indicative of obstacles. Obstacle detection is achieved through image segmentation techniques, while the distance between the setup and the detected obstacles is estimated using a homography-based approach. The proposed algorithm was evaluated on a comprehensive dataset comprising images from three representative scenarios. Experimental results demonstrate the system's effectiveness in reliably detecting obstacles under nighttime conditions and accurately estimating distances.