LEVELS OF DETAIL IN 3D BUILDING RECONSTRUCTION FROM LIDAR DATA
H. Arefi (a) , J. Engels (a) , M. Hahn (a) and H. Mayer (b)
Stuttgart University of Applied Sciences, Stuttgart, Germany
(hossein.arefi, johannes.engels, michael.hahn)@hft-stuttgart.de
(b) Bundeswehr University Munich, Munich, Germany
helmut.mayer@unibw.de
Commission III, WG 4
KEY WORDS: LIDAR, Geodesic Morphology, Building Reconstruction, 3D Modeling, Ridge Line, Approximation, Minimum
Bounding Rectangle, RANSAC, Hough Transform
ABSTRACT:
3D models of buildings are useful for many applications such as urban planning and environmental simulation, cartography, tourism
and mobile navigation. Automatically generating building models in the form of 3D CAD representation is the major part of city
modeling and a challenge for many researchers. Airborne laser-scanning (ALS) results into high-quality geometrical information about
the landscape. It is suitable for the reconstruction of 3D objects like buildings because of its high density and geometrical accuracy.
In this paper a novel approach is proposed for automatically generating 3D building models based on definition of Levels of Detail
(LOD) in the CityGML standard. Three levels of detail are considered in this paper. In the first LOD (LODO), the Digital Terrain
Model extracted from LIDAR data is represented. For this purpose the Digital Surface Model is filtered using geodesic morphology.
A prismatic model containing the major walls of the building is generated to form the LODI. The building outlines are detected by
classification of non-ground objects and the building outlines are approximated by two approaches; hierarchical fitting of Minimum
Boundary rectangles (MBR) and RANSAC based straight line fitting algorithm. LOD2 is formed by including the roof structures into
the model. For this purpose, a model driven approach based on the analysis of the 3D points in a 2D projection plane is proposed. A
building region is divided into smaller parts according to the direction and the number of ridge lines, which are extracted using geodesic
morphology. The 3D model is derived for each building part. Finally, a complete building model is formed by merging the 3D models
of the building parts and adjusting the nodes after the merging process.
1 INTRODUCTION
3D building reconstruction is a challenging problem addressed
by many researchers. Since airborne LIDAR data appeared as a
new data source in remote sensing and photogrammetry many at
tempts were made to model buildings using LIDAR data. LIDAR
combined with aerial images was e.g., used for building recon
struction by (Haala and Anders, 1997, Rotensteiner and Jansa,
2002). The LIDAR data is employed for segmentation of pla
nar faces and the aerial image is involved to improve the quality
of edge segments. The combination of LIDAR data and existing
ground plans was e.g., proposed by (Vosselman and Dijkman,
2001). They employed two strategies for building reconstruction.
The first strategy is based on detection of intersection lines and
height jump edges between planar faces. In second strategy, a
coarse 3D model is refined by analyzing the points that do not
fit well to the coarse model. The first approach which used only
LIDAR data for building reconstruction was presented by (Wei-
dner and Forstner, 1995). They mainly used two types of models;
simple parametric models for buildings with rectangular ground
plans and prismatic models for complex buildings. (Maas, 1999)
developed another model driven approach based on analysis of
invariant moments of the segmented regions to model buildings
in LIDAR image. He assumes that buildings consist of certain
structures such as gable roofs. A prismatic building model based
on edge detection is extracted in (Alharthy and Bethel, 2002).
However, the algorithm is devised for buildings with rectangular
shapes and flat roofs only. A segmentation based approach is pro
posed by (Rottensteiner and Briese, 2002) to find planar regions
which figure out a polyhedral model. Another segmentation ap
proach that uses a TIN structure for the LIDAR surface model
is proposed by (Gorte, 2002). Segments are created by iterative
merging triangles based on similarity measurements. Finally, the
segmented TIN structures are transformed into a VRML model
for visualization.
In this paper a new method is proposed for generating 3D build
ing models in different levels of detail. They follow the standard
definition of the City Geography Markup Language (CityGML)
described in (Kolbe et al., 2005). The CityGML defines five lev
els of detail for multi-scale modeling: LODO - Regional model
contains 2.5D Digital Terrain Model, LODI - Building block
model without roof structures, LOD2 - Building model includ
ing roof structures, LOD3 - Building model including detailed
architecture, LOD4 - Building model including interior model.
Algorithms for producing the first three levels are explained in
this paper. According to above categorization, the first LOD cor
responds to the digital terrain model. An approach based on the
filtering of the non-ground regions uses geodesic reconstruction
to produce the DTM from LIDAR DSM (Arefi and Hahn, 2005,
Arefi et al., 2007b). The LODI level contains a 3D representa
tion of buildings using prismatic models, thus the building roof is
approximated by a horizontal plane. Two techniques are imple
mented for approximation of the detected building outline which
are hierarchical fitting of Minimum Bounding Rectangles and
RANSAC based straight line fitting and merging (Arefi et al.,
2007a). To form the third level of detail (LOD2), a projection
based approach is proposed for reconstructing a building model
with roof structures. The algorithm is relatively fast, because
the 2D data are analyzed instead of 3D data, i.e. lines are ex
tracted rather than planes. The algorithm begins with extracting
the building ridge lines. According to the location and orientation
of each ridge line one parametric model is generated. The mod
els of the building parts are merged to form the complete building
model.
