ROM
'AIWAN
.edu.tw
du.tw
easurement
cen considered
on the practical
images as well
mark which is
ling extraction.
ice plane, or a
By fitting the
> building. The
ate model, (2)
ly edit, and (5)
was developed
(CKU campus.
. 2000), or to
osselman and
Heuvel, 2000,
2001). These
echnology to a
2,
developed for
1s of drawing,
made it being
he increasing
iny researches
j| to extracting
s et al., 1999,
ng and Wang,
rammetry and
| a new term:
ow that using
otogrammetric
ject modelling
(CSG), (2) the
| CAD-based
tool — floating
ents a flexible
| line segment,
ich model is
a set of pose
datum point's
pe parameters
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
change the volume and the outline of the model. From the
traditional photogrammetric point of view, the floating model is
an extension of the floating mark. Instead, it is not only floated
in the object space, but also deformable to fit the outline of the
object. From the model-based building extraction point of view,
floating mark is an exceptional case of floating model without
any shape parameter. The details of floating models are given in
the next chapter.
Comparing to other modelling schemes, such as polyhedral
models, prismatic models, or parameterized polyhedral models,
CSG models scheme do not use one complicated model for a
whole building, but a combination of several volumetric
primitives. À primitive is a predefined simple solid model
which is associated with a number of parameters representing
the shape and pose. The final complete model is composed of a
number of primitives, following the Boolean set operation.
These characteristics make CSG models flexible enough for
modelling most of the buildings, yet still can be simply
described by a small set of rules. Therefore, CSG modeling
scheme is suggested for model-based building extraction by
many pioneers (Braun et al., 1995, Giilch ef al., 1998, Veldhuis,
1998, Ermes ef al., 1999, Tangelder et al., 1999). The proposed
floating model idea complies with the CSG principle. Each kind
of floating models is a primitive. A building may be modelled
by a combination of various floating models.
Model-based building extraction (Sester and Fôrstner, 1989,
Vosselman, 1998, Brenner, 1999, Fischer er al., 1999, Ameri,
2000, Suveg and Vosselman, 2000, Tseng and Wang, 2003)
starts with hypotheses of building model representing a
specified target on the scene, and verifies the compatibility
between the model and the existing image data. Approaches to
model-based building extraction are mostly implemented in a
semi-automatic manner, solving the model-image fitting
problem based on some high-level information given by the
operator. The spatial data of a building object are determined,
when model-image fitting is achieved optimally. Therefore, the
key is the algorithm that is able to determine the pose and shape
parameters of a floating model such that the edge lines of the
wire-frame, as projected into the images, are optimally
coincided with the corresponding edge pixels. It is assumed that
the image orientations are known and the pose and shape
parameters are approximately determined through an interactive
process. To deal with this problem, we proposed a tailored
least-squares model-image fitting algorithm in chapter 3 as a
major component of the building extraction framework.
The proposed semi-automated strategy is shown as figure |.
The basic idea is based on “Human is good at interpretation,
computer is good at calculation.” Therefore, the high-level tasks,
such as primitive selection, approximately fitting, Boolean set
operation, and local modification, are accomplished by operator.
While the low-level task, such as complicated calculation of
optimally fitting, is accomplished by computer programs.
Manual Operations
Model Base ©
Ü
T Em
CSG Boolean
Optimal
Primitive Interactive Operation
n eti -» Approximate HP] Model-image and
Pc es Fitting Fitting Local.
Modification
Figure 1. The proposed semi-automated strategy for model-
based building extraction.
To test the practicabilitv and to evaluate the accuracy, we
designed a computer program for implementing the proposed
procedures. Ten various buildings around the NCKU campus
within four overlapped aerial images were selected as the
experimental objects. Results and accuracy assessment are
given in chapter 4, which shows the practical capability and
potential.
2. BUILDING MODELLING
Building modelling and model-image fitting are two major
issues in model-based building extraction. The issue on building
modelling is how to establish a set of representative and
complete building models. Buildings in Taiwan are highly
diverse in appearance. lt is almost impossible to categorize
them into distinct styles of representative models. However,
there is still regularity inherent in the most building structure,
which allows describing buildings with a small set of rules.
Since the buildings are basically volumetric objects, it is
adequate to modelling by 3D solid entities. By summarizing the
experiences of measuring by floating mark and reviewing the
references of line photogrammetry and CAD-based
photogrammetry, a naval measuring tool of “floating model” is
proposed.
2.1 Floating Models
Traditional photogrammetric mapping systems concentrate on
the accurate measurement of points. The floating mark is a
simple way to represent the position of a point in the space, and
thus, has been served as the only measuring tool on the stereo
plotters up to nowadays. The idea behinds the floating mark is
to depict the intersection V, of the bundle from the projection
centers O, and O,, through the image point v;' and v;" to the
ground, as figure 2 shows. If the conjugate point v;' or v,”
moves along the epipolar line, the intersection point V;
represented by floating mark will raise or sink along the bundle,
seems like "floating" in the object space. This simple
representation of a 3D coordinates has been very useful for
photogrammetric measurement and 2.5D mapping system.
However, the floating mark reaches its limits when the
conjugate points can not be identified due to the occlusions or
interferences from other noises. And with the increasing needs
of 3D object models, point-by-point measurement has been
become the bottleneck of the production.
0, #0,
Figure 2. The concept of floating model.
Line photogrammetry takes the nature of 3D linear feature into
consideration and introduces the 3D line equations as unknowns
instead of 3D coordinates. Take figure 2 for example, operator
