2 THE SYSTEM
This section describes the most important properties of the
system, namely its internal building model, its interaction
capabilities and its features of automation.
2.1 Modeling Buildings
Automatic as well as interactive systems need an internal
model of the objects to be acquired. Buildings show an amaz-
ingly high diversity in structure which is increasing due to
new styles being developed or invented. However, a large
percentage of buildings show regularities which allow to de-
scribe them using a small set of rules. Depending on the
context within an acquisition system we need to distinguish
volumetric, boundary and wire frame type of models.
Volumetric Models Volumetric models are most intuitive
for describing solids. In our context this on one hand eases
interaction. On the other hand, this type of high-level repre-
sentation allows to establish a link between GIS- and CAD-
systems, which seems to be important for many applications.
Many simple buildings can be described using a few param-
eters. Such parametric models can effectively be used for
buildings with flat, gable or hip roof, requiring 3, 4 or 5 pa-
rameters for specifying the form. More general shapes are
prisms with polygonal ground plan and fixed height, which
are frequently found in the central part of cities. The num-
ber of parameters is 2n -- 1 for a ground plan with n corners.
Obviously the ability to model buildings using only these two
types of representation is limited. But combinations of these
appear to be sufficient for modeling a large percentage. This
suggests to immediately use the modeling tool from CAD,
namely Constructive Solid Geometry (CSG) for describing
complex shapes. CSG uses basic parametric shape primitives
and allows to combine them by boolean operations, namely
union, intersection and difference. Complex objects internally
are then represented as a CSG-tree where each node links
two branches of the object with the boolean operation and
a geometric transformation of the reference system of the
branch into a common reference system. The leaves of the
tree are instantiated primitives.
Observe that prisms do not immediately fit into that system,
as they in a weak sense contain free forms, namely the polyg-
onal ground plan.
We use both types of representation. For the CSG-
representation we use as primitives the box, the chock, the
half chock, the cone, the cylinder and the rectangular pyra-
mid. In order to ease acquisition we also have the above
mentioned basic parametric building models as primitives.
Boundary Representation The boundary representation of
parametric and prismatic models are used to support the in-
teraction with the user and for visualization purposes.
This allows to store attributes to all faces of the buildings, es-
pecially surface texture and thus opens the door to photo-true
visualization where the texture can be fused from all images
[Leberl et al., 1994]. Much more, interactive measurements
could be taken at a later stage, here, however, working in
views which not necessarily coincide with those of the origi-
nal images.
We do not provide a complete boundary representation at
the moment, as Computer Graphic packages allow visualiza-
Figure 1: CSG-tree, containing 4 primitive showing the union
of two saddle roofs and two boxes.
<> Glue dist = 2.0 Tree saved.
un
ee EE ee re e
Sh un
—_— qe 24
Sh un
B0 Bo
tion without explicit intersection of the objects. We however
will use the explicit boundary representation for two tasks:
1.) learning regularities at the object for supporting image
analysis and 2.) iconic matching of the final result of the
reconstruction with the content of all images for improving
the accuracy and checking the consistency.
Wire Frame Representation A wire frame representation
is used for interaction. Its model is shown overlayed with the
raster image for interactively fitting the model to the image
content. A hidden line algorithms is running real time in order
to ease cognition of the 3D-structure.
The wire frame model also is used within the model-image
matching procedure based on extracted image edges.
A Sample Representation The example shows the internal
representation for a building with a gable roof. The coordi-
nates of the corners are given in a local coordinate system
in dependence on a set of parameters, here length I, width b
height of building hi, and height of roof hz The coordinates
of the corner points 2 and 8 are given in the local coordinate
system:
X» 0 —1 0 O
Y? 9--o* 0 0 l
Z2 iis 9g '9o 19 b
Xs EI 020 0 0 hı
Ys 1:90 0 ha
Za 00 1
Four additional pose parameters, three shifts and the az-
imuth, are needed to represent the relation of that building
primitive with respect to the world coordinate system.
The model also contains all edges and faces, thus implic-
itly also all relations and constraints necessary to specify the
polyhedral object to be a gable-roof building.
2.2 Interaction
We start with digital or digitized images with known inte-
rior and exterior orientation. A DEM is useful for providing
approximate values.
Extracting CSG-primitives. The 3D-shape extraction of
each single primitive of a CSG-model is performed in several
steps:
1. Choice of the type of boolean operation to previously
extracted primitives, if appropriate, and choice of type
of primitive from a prespecified set.
262
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
Fi
ga
lin
po
