Layer-Wise Floorplan Extraction for Automatic Urban Building Reconstruction.

Area of Science:

• Computer Vision
• Geographic Information Systems
• 3D Modeling

Background:

• Urban digitization and realistic visualization require accurate building geometry reconstruction.
• Existing methods may struggle with complex urban structures and data noise.

Purpose of the Study:

• To propose a novel automatic method for recovering urban building geometry from 3D point clouds.
• To develop a robust approach suitable for common urban buildings composed of planar polygons aligned with gravity.

Main Methods:

• Formulating building reconstruction as an energy minimization problem within a Markov Random Field (MRF) framework.
• Slicing point clouds along the gravity direction to reconstruct floorplans, extracting dominant shapes, and propagating them using MRF.
• Designing a novel string-based distance metric for accurate floorplan comparison and propagation.
• Incorporating an image-based editing method for detailed window structure recovery.

Main Results:

• The proposed method effectively reconstructs urban building geometry from 3D point clouds.
• Experimental results on synthetic and real datasets validate the method's effectiveness and robustness.
• The approach successfully handles buildings with piecewise constant shapes along the gravity direction.

Conclusions:

• The developed automatic method offers a significant advancement in urban building reconstruction.
• This technique contributes to improved urban digitization and realistic 3D visualization.
• The MRF-based approach with novel floorplan distance measurement ensures accurate geometry recovery and detailed structure reconstruction.