Repository logo

Automated Generation of Railway Track Geometric Digital Twins (RailGDT) from Airborne LiDAR Data

Accepted version


Conference Object

Change log


Ariyachandra, Mahendrini Fernando 


Automated generation of railway track geometric digital twins (RailGDT) from airborne LiDAR data is an unresolved problem. Currently, this onerous manual procedure counteracts the expected benefits of the resulting RailGDT. State-of-the-art methods provided promising results, but are unable to generate RailGDTs over kilometres with complex railway geometries without forfeiting precision and manual cost. The challenge that this paper address is how to efficiently minimise manual cost for generating RailGDTs such that the benefits provide even greater compared to the initial investment in RailGDTs. We tackle this challenge by leveraging the highly standardised nature of railways. The method restricts the search region and segments track elements given their locations relative to masts, using an extended RANSAC algorithm. Next, it converges segmented point clusters with various pre-assembled track element profiles to obtain RailGDTs. Experiments on 18 km datasets yield 95% and 98% average F1 scores for rail and trackbed point cluster segmentation. The RailGDT accuracy is 3.4 cm and 2.7 cm RMSEs for rails and trackbeds.



Journal Title

EG-ICE 2021 Proceedings: Workshop on Intelligent Computing in Engineering

Conference Name

28th International Workshop on Intelligent Computing in Engineering (EG-ICE 2021)

Journal ISSN

Volume Title


Universitätsverlag der TU Berlin, 2021


All rights reserved
Leverhulme Trust (IAF-2018-011)
Engineering and Physical Sciences Research Council (EP/S02302X/1)
Australian Research Council (DP170104613)
Cambridge Commonwealth, European & International Trust Bentley Systems UK Plc