Jung, Jaewook (Hitachi), Ghorbanalivakili, Mohammadjavad (York University), Sohn, Gunho (York University)

Multi-Railway Track and Switch Region Recognition Using Mobile Laser Scanning Data

Scheduled for presentation during the Regular Session "S37c-Reliable Perception and Robust Sensing for Intelligent Vehicles" (FR-LA-T37), Friday, November 21, 2025, 16:00−16:20, Coolangata 1

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Railway infrastructure is a vital component of modern transportation systems. However, today’s railway equipment inspections primarily rely on labor-intensive fieldwork and error-prone human visual analysis. In this study, we present an automated system for the recognition of railway assets using mobile laser scanning data. Our method simultaneously detects rail tracks and localizes switch regions, identifying switch region orientation (right/left), status (open/closed), and overlap type (merge/split). This integrated railway tracing and switch recognition method is built on binary multiscale template matching within a Kalman filtering framework. To trace the railway trajectories, the Kalman filter predicts a new state vector that consists of the position and orientation of the rail track window. This prediction which is based on the previous state vector is then updated using the observed track points already classified in a local track window using the Gaussian mixture model clustering method. To detect multi-tracks and determine appropriate observations for the targeted rail track, we employ minimum description length technique. Lastly, to recognize the switch type, template matching is applied by considering the similarity between a multi-track region and the templates as well as track design constraints observed in the sequential multi-track regions. We test our algorithm on a densely-annotated private benchmark of 3D point cloud data captured from a large-scale railway network. Experiments on this dataset show that the proposed method can robustly produce accurate rail track vectors (99.46% recall) and recognize switch types (97% success rate).