Automating maintenance of road Geometric Digital Twins through single scan instance aware point cloud change retrieval

Proactive road maintenance extends asset lifespan, enhances safety and reduces downtime. However, costly reactive maintenance becomes necessary without up-to-date and well-structured data. While Geometric Digital Twins (GDT) offer digital replicas of physical structures that can be used to automate maintenance activities, no automated tools are currently available for GDT’s upkeep. This paper addresses this issue and proposes a method with a multi-step pipeline for detecting changes, matching instances, identifying newly added objects and applying these changes to the GDT 3D model using point clouds. Our methods, namely Iterative Change Refinement using a single labelled scan and Dual Instance Aware Change Retrieval using two scans, achieve a 0.89-0.97 F1 score in change detection, and our holistic pipeline results in less than 0.1° rotation angle and 0.01 m translation mean squared errors. This pipeline automates the process of GDT maintenance, making such digital twins viable and practically applicable to the industries. • Automatic maintenance of 3D road geometric digital twins. • Dual Instance Aware Change Retrieval achieves 0.95-0.97 F1 score in change detection. • Iterative Change Refinement using a single labelled point cloud scan achieves a 0.89 F1 score for change detection. • Our holistic pipeline results in less than 0.1° rotation angle and 0.01 m translation MSE.