SEMI-AUTOMATED CSG MODEL-BASED BUILDING EXTRACTION FROM
PHOTOGRAMMETRIC IMAGES
S. D. Wang“, Y. H. Tseng ^
Department of Geomatics, National Cheng Kung University, No. 1 University Road, Tainan 70101, TAIWAN
2 Bh. D. Candidate, TEL: +886-6-2370876 ext.852, FAX: +886-6-2375764, E-Mail: sendo@sv.ncku.edu.tw
° Professor, TEL: +886-6-2757575 ext. 63835, FAX: +886-6-2375764, E-Mail: tseng@mail.ncku.cdu.tw
Commission III, WG III/4
KEY WORDS: Digital, Photogrammetry, Semi-automation, Building, Extraction, Three-dimensional, Modelling, Measurement
ABSTRACT:
Using pre-defined models to extract spatial information of the building, called model-based building extraction, has been considered
as a convincible approach to improve the existing photogrammetric techniques. However, there is still a bottleneck on the practical
efficiency and accuracy. This paper proposed a semi-automated approach to extracting buildings from multiple aerial images as well
as close-range images by a tailored least-squares model-image fitting (LSMIF) algorithm. Compare to the floating mark which is
used by conventional photogrammetric techniques, we proposed a naval measuring tool of “floating models” for building extraction.
The floating model is an abstract representation of the real object, which can be a point, a line segment, a surface plane, or a
volumetric model. Each floating model is associated with a set of pose parameters and a set of shape parameters. By fitting the
model to images, these parameters will not only reveal the location of the building but also describe the shape of the building. The
semi-automated strategy for building extraction includes following five steps: (1) manually select an appropriate model, (2)
manually locate and adjust the model for approximate fitting, (3) automatically compute the optimal fittin
compose the fitted models according to the Constructive Solid Geometry (CSG). An ad hoc computer program which was developed
to implementing the proposed semi-automated approach was tested by extracting 10 selected buildings around the NCKU campus.
The accuracy achieved was evaluated by comparing the roof corner coordinates with manual measurements.
1. INTRODUCTION
In response to the development of 3D City Spatial Information
Systems for urban planning and management, acquisition of 3D
data of city objects has increasingly become an important topic.
This tendency leads to intense research activities aiming for
automated or semi-automated building extraction from digital
aerial images in both the photogrammetry and the computer
vision communities (Mohan and Nevatia, 1989, Braun et al,
1995, Englert and Gülch, 1996, Lang and Forstner, 1996,
Vosselman and Veldhuis, 1999, Grün, 2000). While the task of
building extraction may differ in terms of image data type and
scale, object complexity, required level of detail, and type of
product, the common process sequence would be: detection,
reconstruction, and attribution. Various approaches have been
implemented with emphasis on more or less automation with
respect to the process sequence.
Conventional photogrammetry concentrates on the accurate 3D
coordinate measurement of points. The automated measuring
systems set up by image matching algorithms are still based on
the point-to-point correspondence. However, linear feature
contains more geometric and semantic information than point. It
is also casier to be extracted from the photogrammetric images.
Since the last decade, scholars and experts have been exploring
the methods that linear feature takes place of point for solving
photogrammetric problems (Schenk ef al., 1991, Li and Zhou,
1994). The researches show that linear features can be used to
determine image orientation by space resection (Petsa and
Patias, 1995, van den Heuvel, 1997, Kerschner, 1998, Hrabacek
and van den Heuvel, 2000, Smith and Park, 2000), or to
measure object by model-image fitting (Vosselman and
Haralick, 1996, van den Heuvel, 1999, van den Heuvel, 2000,
Heuel and Forstner, 2001, Zhou and Li, 2001). These
innovative researches lead the photogrammetric technology to a
new stage which is called “Line Photogrammetry”.
Although the CAD system is not initially developed for
photogrammetric purpose, its powerful functions of drawing,
manipulating, and visualizing 2D objects have made it being
widely used with photogrammetric systems. The increasing
demands of object’s 3D models encourage many researches
toward using 3D CAD models as a modelling tool to extracting
objects from image data (Das ef al., 1997, Ermes et al., 1999,
Boehm et al., 2000, van den Heuvel, 2000, Tseng and Wang,
2002). This trend towards integration of photogrammetry and
CAD system in the algorithmic aspect creates a new term:
"CAD-based Photogrammetry”. Researches show that using
CAD models does increase the efficiency of photogrammetric
modelling by two reasons: (1) the advanced object modelling
techniques such as Constructive Solid Geometry (CSG), (2) the
incorporation of geometric object constraints.
Inspired by the line photogrammetry and CAD-based
photogrammetry, we proposed a naval measuring tool — floating
model — in this paper. The floating model represents a flexible
entity floating in the 3D space. It can be a point, a line segment,
a surface plane, or a volumetric model. Each model is
associated with a set of shape parameters and a set of pose
parameters. The pose parameters determine the datum point's
position and the rotation of the model. The shape parameters
g, (4) manually edit, and (5)
Interne
change
traditic
an exte
in the
object.
floatin,
any sh:
the ne
Compe
models
CSG 1
whole
primiti
which
the sha
numbe
These
modell
descrit
schem
many |
1998, .
floatin
of floa
by a cc
Model
Vossel
2000,
starts
specifi
betwee
model-
semi-a
problei
operat
when 1
key is
param.
wire-fi
coincic
the ir
param.
proces
least-si
major
The pi
The b:
compu
such a
operat;
While
optima
Figure
