BUILDING EXTRACTION FROM MULTIPLE DATA SOURCES: A DATA FUSION
FRAMEWORK FOR RECONSTRUCTION OF GENERIC MODELS
K. Khoshelham
Dept. of Land Surveying and Geo-Informatics, The Hong Kong Polytechnic University, Hong Kong
Kourosh.k@polyu.edu.hk
Commission III, WG III/4
KEY WORDS: Building Extraction, Fusion, Automation, Modelling, Aerial image, Laser scanning, GIS
ABSTRACT:
Automated building extraction from multi-source data has attracted great attention in recent years. This paper presents an approach to
automated building extraction by fusion of image data, height data and 2D ground plans. In this approach buildings are detected
using ground plans and height data. A split-and-merge process is applied to fuse image and height data and derive the parametric
forms of roof planes. Vegetation regions are identified and discarded using the image information in red and infrared channels. Walls
are reconstructed as vertical planes upon the ground plan. The model planar faces are finally intersected and resulting plane patches
are assembled together to form a generic polyhedral model. Results of the experimental testing indicate the promising performance of
the proposed approach in automatic detection and reconstruction of buildings.
1. INTRODUCTION
Automated building extraction has been a challenging problem
in the past two decades. Automated approaches that work solely
based on a single source of data, suffer from the lack of
robustness due to complexities in data as well as in buildings.
Therefore, in recent years research efforts have been focused on
automated approaches that make use of data from multiple
sources. This paper presents a framework for fusion of available
data sources that can be used in an automated system for
extraction of building objects.
Various types of data from different sources have been used for
automated extraction of buildings. Aerial images are the most
widely used data. Single aerial images have been used for
automated detection and reconstruction of buildings with simple
models (Huertas et al., 1993; Lin et al., 1995; Lin and Nevatia,
1996; Nevatia et al., 1997; Shufelt and McKeown, 1993). For
reconstruction of more complex buildings, stereo and multiple
overlap aerial images have attracted greater attention (Baillard
et al., 1999; Bignone et al., 1996; Dang et al., 1994; Fischer et
al., 1998; Fua and Hanson, 1991; Henricsson, 1998; Henricsson
and Baltsavias, 1997; Herman and Kanade, 1986; Jaynes et al.,
2003; Kolbe, 1999; Moons et al., 1998). Image data from other
sources have not been suitable for building reconstruction.
Remotely sensed images from satellites are of relatively low
ground resolution and, therefore, can only be used for detection
of buildings. Close range images are, on the other hand, too
detailed and can be used to map textures onto final
reconstructed models.
Height data is another widely used type of data. Digital surface
models (DSMs) from aerial laser scanning systems have been
used in a number of approaches (Brunn and Weidner, 1997;
Maas, 1999; Vosselman, 1999; Weidner and Forstner, 1995).
Range data from terrestrial laser scanners, however, has not
been proved useful for automated building extraction.
Automated building extraction from both image and height data
encounters a number of complexities. Image data often suffers
from noise, low contrast, shadow and occlusion; hence, features
extracted from images are incomplete and uncertain. Height
data is of relatively low resolution, which makes the extraction
of building boundaries difficult. These complexities have led
the research efforts toward methods that combine data from
multiple sources. In recent years a number of methods have
been developed for fusion of image and height data (Amer,
2000; Cord et al., 2001; Jaynes et al., 2003; Rottensteiner and
Jansa, 2002), although height data in some of these methods has
been generated from image data using matching techniques.
Fusion of image data and 2D ground plans has also appeared in
a number of works (Haala and Anders, 1996; Jibrini et al., 2000;
Pasko and Gruber, 1996; Suveg and Vosselman, 2004). In
another fusion strategy, Haala and Brenner (1998) used DSM
and 2D ground plans in an approach to automated building
extraction.
Despite the great deal of research that has been carried out on
automated building extraction, still the role of multi-source data
and fusion strategies has not been completely explored. The
objective of this paper is to develop a framework for fusion of
available data sources that can be used for automated extraction
of buildings. The proposed fusion framework combines aerial
images in colour and infrared channels, DSM and DTM from
aerial laser scanner and 2D ground plans from a GIS database in
an approach to automated extraction of buildings. In this
approach ground plans are used to detect buildings in the scene
and reconstruct the walls. Image data in red and infrared
channels are used to identify and remove vegetation regions.
Roof planes are reconstructed by exploiting information from
image, DSM and DTM. Generic polyhedral models are finally
formed by assembling reconstructed walls and roof planes.
The paper is structured in 6 sections. In section 2 an overview
of the proposed fusion strategy is presented. Reconstruction of
walls and roof planes are described in section 3. Section 4
discusses the reconstruction of generic models using a plane
patch reconstruction technique. Experiments and results are
shown in section 5. Conclusions are made in section 6.
Intert
—
2 7?
]
The f
build
proje:
third
the ht
Havii
is bas
Thes
patch
mode
heigh
starts
Heigl
regio
faces
in wi
neigh
Vege
an N
Ever
over
height
“other
than
large
Wall
every
heigt
bene:
faces
inter:
a cor
sorte
final
grapl
The
form
build
types
appr
boun
Neve
Cross
Cons
way,
(200
parts
recoi
then
3
Bui l
heig
and |
form
para
The
deta
