Leveraging railway topology to automatically generate track geometric information models from airborne LiDAR data

Geometric information modelling from point cloud data (PCD) is a fundamental step of the digital twinning process for rail infrastructure. Currently, this onerous procedure outweighs the anticipated benefits of the resulting model and expends 74% of the modellers' effort on converting PCD to a model. The cost of the resulting geometric information models (GIM) can be reduced by automating the modelling process. State-of-the-art methods cannot offer large-scale GIM generation required over kilometres without forfeiting precision and manual cost. This paper addresses the challenge of achieving such automation by leveraging the highly standardised topology of railways to automatically generate GIMs of rail track structures. The method first automatically segments rails and track beds, delivering labelled point clusters of track structure elements. Next, it converges the segmented rails with pre-defined parametric assemblies of different rail profiles and uses a mesh-based approach to reconstruct the geometry of the track bed, delivering industry foundation classes (IFC) files of railway track structure elements. Experiments on 18.5 km railway PCDs yielded an average segmentation of 98.1% and 94.9% F1 scores and overall modelling accuracy of 3.5 cm and 2.8 cm root mean square error (RMSE)s for rails and track beds. The proposed method can realise an estimated time savings of 88.9% without needing any manual inputs.