Subproject B01

Adaptive geometric-semantic multi-LOD construction model of road infrastructure as Cyber-Physical System (CPS)

For a consistent description of the shape and location of the road infrastructure, an adaptive geometric-semantic building model (GSM) with several abstraction levels is developed, which enables the spatially multi-scale geometric representation of the real existing building object and serves as a homogeneous spatial reference for all geometry-related questions in the SFB/TRR 339. As a georeference, the model represents the connection between the real street and its digital shadow. The automatic model derivation is based on a combination of algorithmic geometry, knowledge-based approaches and machine learning methods. The GSM will represent the interface to Building Information Modeling (BIM) and Geospatial Information Modeling (GIM), where GSM are a central component. The GSM data can be used in different modeling depths (Level of Detail, LOD) for simulations, visualizations and other purposes in order to work on questions of the involved SP. Besides the spatial resolution, the geometric-semantic reliability plays an important role. For this purpose, a novel combined quality measure is developed, which allows a fuzziness quantification of the GSM to the required LOD. For the SP, primarily point clouds from laser scans and images, digital planning models (CAD/BIM) as well as additional information from GIS are used.

Workflow of GSM creation: First, the point cloud is segmented to reconstruct spatial bodies for the selected LOD. These are classified using a hierarchical model. Any blurring is quantified using a quality measure.
Influence of synthetic training data from different sources on the validation result with real data.
Influence of pre-segmentation on the resulting sharpness of detail of the geometry construction

Project Participants

Univ.-Prof. Dr.-Ing. habil.
Jörg Blankenbach
Subproject Manager
David Crampen
Scientific Researcher
Marius Laska
Associated Member
Jan Martens
Associated Member

Publications of the Subprojects


      You are using an outdated browser. The website may not be displayed correctly. Close