Advances in the Multimodal 3D Reconstruction and Modeling of Buildings

Information

• Publication Type: PhD-Thesis
• Workgroup(s)/Project(s):
  • Rendering and Modeling
  • VRVis Competence Center
• Date: June 2018
• Date (Start): 2010
• Date (End): 2018
• Open Access: yes
• 1st Reviewer: Elmar Eisemann
• 2nd Reviewer: Hannes Kaufmann
• Rigorosum: 23. October 2018
• First Supervisor: Michael Wimmer

Abstract

Driven by the need for faster and more efficient workflows in the digitization of urban environments, the availability of affordable 3D data-acquisition systems for buildings has drastically increased in the last years: Laser scanners and photogrammetric methods both produce millions of 3D points within minutes of acquisition time. They are applied both on street-level as well as from above using drones, and are used to enhance traditional tachymetric measurements in surveying. However, these 3D data points are not the only available information: Extracted meta data from images, simulation results (e.g., from light simulations), 2D floor plans, and semantic tags – especially from the upcoming Building Information Modeling (BIM) systems – are becoming increasingly important. The challenges this multimodality poses during the reconstruction of CAD-ready 3D buildings are manifold: Apart from handling the enormous size of the data that is collected during the acquisition steps, the different data sources must also be registered to each other in order to be applicable in a common context – which can be difficult in case of missing or erroneous information. Nevertheless, the potential for improving both the workflow efficiency as well as the quality of the reconstruction results is huge: Missing information can be substituted by data from other sources, information about spatial or semantic relations can be utilized to overcome limitations, and interactive modeling complexity can be reduced (e.g., by limiting interactions to a two-dimensional space). In this thesis, four publications are presented which aim at providing freely combinable “building blocks” for the creation of helpful methods and tools for advancing the field of Multimodal Urban Reconstruction. First, efficient methods for the calculation of shadows cast by area light sources are presented – one with a focus on the most efficient generation of physically accurate penumbras, and the other one with the goal of reusing soft shadow information in consecutive frames to avoid costly recalculations. Then, a novel, optimization-supported reconstruction and modeling tool is presented, which employs sketch-based interactions and snapping techniques to create water-tight 3D building models. An extension to this system is demonstrated consecutively: There, 2D photos act as the only interaction canvas for the simple, sketch-based creation of building geometry and the corresponding textures. Together, these methods form a solid foundation for the creation of common, multimodal environments targeted at the reconstruction of 3D building models.

Additional Files and Images

Weblinks

• Entry in reposiTUm (TU Wien Publication Database)

BibTeX

@phdthesis{schwaerzler_2018_phd,
  title =      "Advances in the Multimodal 3D Reconstruction and Modeling of
               Buildings",
  author =     "Michael Schw\"{a}rzler",
  year =       "2018",
  abstract =   "Driven by the need for faster and more efficient workflows
               in the digitization of urban environments, the availability
               of affordable 3D data-acquisition systems for buildings has
               drastically increased in the last years: Laser scanners and 
               photogrammetric methods both produce millions of 3D points
               within minutes of acquisition time. They are applied both on
               street-level as well as from above using drones, and are
               used to enhance traditional tachymetric measurements in
               surveying. However, these 3D data points are not the only
               available information: Extracted meta data from images,
               simulation results (e.g., from light simulations), 2D floor
               plans, and semantic tags – especially from the upcoming
               Building Information Modeling (BIM) systems – are becoming
               increasingly important. The challenges this multimodality
               poses during the reconstruction of CAD-ready 3D buildings
               are manifold: Apart from handling the enormous size of the
               data that is collected during the acquisition steps, the
               different data sources must also be registered to each other
               in order to be applicable in a common context – which can
               be difficult in case of missing or erroneous information.
               Nevertheless, the potential for improving both the workflow
               efficiency as well as the quality of the reconstruction
               results is huge: Missing information can be substituted by
               data from other sources, information about spatial or
               semantic relations can be utilized to overcome limitations,
               and interactive modeling complexity can be reduced (e.g., by
               limiting interactions to a two-dimensional space). In this
               thesis, four publications are presented which aim at
               providing freely combinable “building blocks” for the
               creation of helpful methods and tools for advancing the
               field of Multimodal Urban Reconstruction. First, efficient
               methods for the calculation of shadows cast by area light
               sources are presented – one with a focus on the most
               efficient generation of physically accurate penumbras, and
               the other one with the goal of reusing soft shadow
               information in consecutive frames to avoid costly
               recalculations. Then, a novel, optimization-supported
               reconstruction and modeling tool is presented, which employs
               sketch-based interactions and snapping techniques to create
               water-tight 3D building models. An extension to this system
               is demonstrated consecutively: There, 2D photos act as the
               only interaction canvas for the simple, sketch-based
               creation of building geometry and the corresponding
               textures. Together, these methods form a solid foundation
               for the creation of common, multimodal environments targeted
               at the reconstruction of 3D building models.",
  month =      jun,
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology ",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2018/schwaerzler_2018_phd/",
}