Abstract:Creating geometric digital twins (gDT) for as-built roads still faces many challenges, such as low automation level and accuracy, limited asset types and shapes, and reliance on engineering experience. A novel scan-to-building information modeling (scan-to-BIM) framework is proposed for automatic road gDT creation based on semantically labeled point cloud data (PCD), which considers six asset types: Road Surface, Road Side (Slope), Road Lane (Marking), Road Sign, Road Light, and Guardrail. The framework first segments the semantic PCD into spatially independent instances or parts, then extracts the sectional polygon contours as their representative geometric information, stored in JavaScript Object Notation (JSON) files using a new data structure. Primitive gDTs are finally created from JSON files using corresponding conversion algorithms. The proposed method achieves an average distance error of 1.46 centimeters and a processing speed of 6.29 meters per second on six real-world road segments with a total length of 1,200 meters.

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve The paper aims to address multiple challenges encountered in the creation of existing road geometric digital twins (gDT), including low automation, insufficient accuracy, limited asset types and shapes, and reliance on engineering experience. Specifically, the paper proposes an automatic scan-to-BIM framework based on semantically annotated point cloud data (PCD) for creating geometric digital twins of roads. The framework mainly addresses the following issues: 1. **Low automation**: Existing methods require a significant amount of manual intervention when extracting geometric information of road assets, leading to inefficiency. 2. **Insufficient accuracy**: Existing methods have large errors when creating geometric digital twins, failing to meet high-precision requirements. 3. **Limited asset types and shapes**: Existing methods mainly focus on extracting road assets with regular shapes (such as pavements and lane markings), while providing insufficient support for assets with arbitrary shapes (such as road signs, street lights, and guardrails). 4. **Reliance on engineering experience**: Existing methods rely on specific engineering experience and standards when processing input data due to limited or low-quality information, which restricts their flexibility and accuracy in practical applications. To solve these problems, the paper proposes a new framework that achieves the following steps: 1. **Instance segmentation**: Using density clustering algorithms (such as DBSCAN) to segment point cloud data with the same semantic labels into multiple spatially independent instances. 2. **Geometric information extraction**: Developing specific algorithms for different types of assets (planar, pole-like, and guardrails) to extract their geometric information. For example, for pole-like assets, using a deep learning part segmentation model to further segment them into specific parts. 3. **Geometric information storage**: Storing the extracted geometric information in a structured and compressed manner in JSON files, rather than directly converting it into specific 3D/BIM file formats. 4. **Geometric digital twin creation**: Developing corresponding conversion algorithms to construct the original geometric digital twin representation from the extracted and stored geometric information. The framework was validated on six actual road segments with a total length of 1200 meters, achieving an average distance error of 1.46 centimeters and an average processing speed of 6.29 meters per second.

Scan-to-BIM for As-built Roads: Automatic Road Digital Twinning from Semantically Labeled Point Cloud Data

Automatic Geometric Digital Twin of Box Girder Bridge Using a Laser-Scanned Point Cloud

A Framework for Digital Twin of Civil Infrastructure— Challenges & Opportunities

Automated Generation of FE Model for Digital Twin of Concrete Structures from Segmented 3D Point Cloud

Semi-automated Generation of Geometric Digital Twin for Bridge Based on Terrestrial Laser Scanning Data

Automated geometric digital twinning of bridges from segmented point clouds by parametric prototype models

Automatic generation of structural geometric digital twins from point clouds

Building Digital Twins of Existing Highways Using Map Data Based on Engineering Expertise

Automated Reconstruction of Parametric BIM for Bridge Based on Terrestrial Laser Scanning Data

Semi-automated generation of parametric BIM for steel structures based on terrestrial laser scanning data

Fast and Accurate Generation Method of Geometric Digital Twin Model of RC Bridge with Box Chambers Based on Terrestrial Laser Scanning

Automatic reconstruction of as-built building information models from laser-scanned point clouds: A review of related techniques

SCAN-VS-BIM FOR REAL-TIME PROGRESS MONITORING OF BRIDGE CONSTRUCTION PROJECT

Feature-constrained automatic geometric deformation analysis method of bridge models toward digital twin

Automated BIM Reconstruction of Full-Scale Complex Tubular Engineering Structures Using Terrestrial Laser Scanning

Framework for Geometric Information Extraction and Digital Modeling from LiDAR Data of Road Scenarios.

A parametric approach towards semi-automated 3D as-built modeling

Improving completeness and accuracy of 3D point clouds by using deep learning for applications of digital twins to civil structures

Computer Vision Enabled Building Digital Twin Using Building Information Model

3D POINT CLOUD TO BIM: AUTOMATED APPLICATION TO DEFINE IFC ALIGNMENT AND ROADWAY WIDTH ENTITIES FROM MLS-ACQUIRED LiDAR DATA OF MOUNTAIN ROADS