International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
Cartographic Generalization
wris files »
v wris files VRML
Representation ++ — — — — — Result of the
Classify objects
Model yo» generalization
Classified user process
objects 1! answer of questions A
v selection : 1! proposed by expert application of
Expert ; | . .System |. operators
System Er ru
Primary and Y operators |
Important > Second indeterminate Expert
secondary objects Representation Model objects » System
Figure 2: Generalization 3D system architecture.
objects, such as rivers and mountains. In the second are
the medium-sized objects, as buildings and houses. The
remaining objects are classified in the third category. The
first category deals with geographic modelling, while the
two others with urban modelling (Frery and Kelner, 2002).
This classification is important because the operators can
be applied differently according to the category.
The expert system with knowledge base about object selec-
tion works in the secondary category selecting the objects
in agreement with the theme, e.g. tourism. Nothing is done
onthe others categories. A simple rule for objects selection
is shown in Table 1: the system looks for keywords in the
object name, as “museum” and “restaurant”, and verifies if
there are other objects near (the user gives the radius).
Table 1: Rule for secundary object selection.
| | If there is keyword in object name or there are
no others secondary objects nearby then Select
object
The Second Representation Model is similar to the first, but
with less objects. The expert system using the knowledge
base about the application of operators applies the opera-
tors to the virtual world. The system defines three levels of
distance: LOD1, LOD2 and LOD3 that will be employed at
distances defined by the user. Table 2 presents the LODs
and their relation with the generalization operators.
Table 2: Main rules for operators application.
1 | If LOD1 <> 0 then apply the simplification opera-
tor
2 | If apply the simplification operator and simplify
primitives then select the object category
3 | If apply the simplification operator e simplificar
IndexedFaceSet then select the object category
4 | If simplify IndexedFaceSet then select the
IndexedFaceSet algorithm simplification
5 | If LOD2 <> 0 then apply the smoothing operator
6 | If apply the smoothing operator then select the ob-
ject category
7 | If LOD3 <> 0 then apply the simbolization opera-
tor
8 | If apply the simbolization operator then select the
object category
4.1 The Implemented Operators
This section presents details of the implementation of gen-
eralization operators in the context of virtual reality. Our
target is system validation, not the implementation of all
the operators. The operators OP8, OP9 and OP10 are pro-
vided in VRML through the Transform node. The imple-
mented operators were simplification, smoothing and sym-
bolization:
Simplification: composed of two algorithms: primitive
simplification and IndexedFaceSet simplification.
The first is responsible for simplifying VRML primi-
tives: box, sphere, cone and cylinder. A VRML
primitive can be built with many faces; for instance,
a sphere can be rendered with sixty faces requiring
computational resources. This algorithm produces a
simplified version of each primitive by projecting it
onto a convenient plane. The new flat object, built
as an IndexedFaceSet, inherits the properties of the
original primitive, e.g. colour, texture and size. Sphe-
res become circles, cones become triangles and boxes
rectangles. Figure 4.1, left, presents an object built
with VRML primitives and, to the right, the result
of the simplification primitive algorithm. They look
alike from a certain distance.
Objects built with IndexedFaceSet have, in most of
cases, many, even millions of faces. Many of these
objects are the result of exporting from 3D CAD plat-
forms, and they are comprised of triangles. Among
the papers with mesh triangle simplification one can
cite (Vieira et al., 2003, Hoppe, 1996, Guéziec et al.,
1999). The IndexedFaceSet algorithm simplifica-
tion reduces the number of faces of the original ob-
ject.
Smoothing: this operator works on textures with image
processing techniques. The target is to get new simi-
lar smaller textures in two steps: applying a low-pass
filter (Lim, 1989) to blur the image and then sampling
it. Figure 4 presents an example of this operator; the
image to the left is the original image, top right is
the blurred one and bottom right is the subsampled
one. Their sizes are, respectively, 118kB, 52kB and
12kB. The subsampling rate is 1 + 3.
Symbolization: this operator changes the objects for sym-
bols which, in turn, are textures over single-faced In-
dexedFaceSets. Each texture is related with a key-
word, and the system looks for keywords in the object
name; if it finds a texture with the same keyword of
the object, a symbol is created. Figure 5 (left) presents
an object called “Statue” built with nine box primi-
tives, two spheres and three IndexedFaceSets (each
one with million of points); it has 191kB and was in-
serted in the system with the keyword “statue”. The
corresponding symbol is shown right top, and right
bottom its visualization from some distance. The sym-
bol requires only 3kB.
The result ofthe generalization process is stored into VRML The system was tested on a large virtual world depicting
files. The system was development in Java.
the historical quarter of the city of Recife (PE, Brazil).
202
Inter
Nav
quir
Figi
The
exib
wor
5
This
the 1
of v
tran
forn
for €
REI
Bou
els.
berg
Coh
200(
Proc
Con
Non
Cros
A.
Teck
