ground surface, i.e. the buildings “sink” in the
terrain. However, this approach shows several
advantages:
e The walls of the building remain rectangular which
facilitates the texturing process.
e Geometric objects remain easy objects for a
manipulation. They can be removed
e Since the building boxes are set out at the point
with minimum height, the
e correspondence between the surface and each
building is ensured. There are not occlusions and
holes occurred around the buildings.
e Rectangularity of the buildings allows a
parametric description of the buildings.
e Detailed attribute information can be embedded
to each of the separate objects (buildings, man-
made objects, etc), which will allow queries about
the model to be carried out.
4.Test area and future work
The ongoing work is based on two test areas from
the central part of Enschede, The Netherlands and
Graz, Austria.. The first data set is from relatively
flat area and is used for investigation of
triangulation and re-triangulation of the surface. The
second data set is from that part of Graz which is
dominated by the Schlossberg. This hill offers a
very rough surface and numerous buildings,
pathways, retaining walls and various other objects.
This data are utilized for texturing and for studying
the way of merging the surface features and terrain
mesh.
Various questions have to be answered in the future
work. The construction of the 3D model still involves
a lot of manual work. The most important directions
for further explorations and developments are:
1. An algorithm for automatic modelling of the
slopes of the surface objects should be
developed. All the slopes along passways and
patches are created manually in the examples
presented in this paper.
2. Algorithms for automatic creation of different
levels of detail for fast visualization have to be
created.
The concept of LOD comprises both aspects of 3D
visualization, i.e. geometry and photo texture. The
geometry in urban areas consists of objects which
are described by set of triangles (surface),
quadrangles (walls of the buildings, etc) and multi-
edges polygons (balconies, ornaments, etc).There
are algorithms developed already for LOD for a
surface presented by grid structure (cf. [5]),
however, still there is no worked out concept for
geometric objects (buildings) and surface features
(passways, etc) laying on the ground. The
complexity of the 3D data elements yields the
following questions:
- How many levels of detail are reasonably to
314
create? The expectations are four LODs for the
building, 3 for the surface and three LODs for the
photo texture.
- What kind of data structure should be used for
storing the different LOD? The research work in this
area gives some possible solution like quadtree (cf.
[5]), BSP (cf. [9]) or R-tree (cf. [10]). In any case
the tree-structure is considered to be the most
suitable data structure for visualization.
- What kind of thresholds should be used for
choosing the necessary LOD during the process of
visualization? Since the thresholds for switching the
LOD might be different for geometry and texture, a
correspondence amongst the LOD must be
established. The distance from the viewer is a
compulsory threshold for both geometry and
texture. Another threshold could be the current
camera position and roughness of the relief.
- How to automate the process of creating different
LODs? The construction of different LOD seems to
be a part of the work which will involve a lot of time
and efforts. Special investigations should be carried
out to determine the necessary techniques. The
basic questions is what criterions should be chosen
for feature generalizations.
References
1. Gruber M, S. Meissl, R. Bohm, Das
Dreidimensionale Stadmodell Wien, Erfahrugen
aus einer Vorstudie, VGI, 1+2/95, Wien 1995
2. Gruber M., M. Pasko, F. Leberl, Geometric
versus texture detail in 3D Models of Real World
Buildings, in Automatic Extraction of Manmade
Objects from Aerial and Space Images, Monte
Verita, Birkhauser Verlag, Basel 1995
3. Heler M., Triangulation algorithms for adaptive
terrain modelling, proceedings of the 4th
International Symposium on Spatial Data
Handling, Zurich, Switzerland, 1990 , pp. 163-174
4. Leberl F., M Gruber, P. Uray, F. Madritsch,
Trade-offs in the Reconstruction and Rendering
of 3-D objects, Mustererkennung'94, Wien 1994
5. Lindstrom P., D. Koller, L. Hodges, W.
Ribarsky, N Faust, G. Turner, Level of Detail
Management for Real-Time Rendering of Photo
Textured Terrain, Technical report 95-14, Georgia
Institute of Technology, 1995
6. Lindstrom P., D. Koller, W. Ribarsky, L.
Hodges, Real-Time, Continuous Level of Detail
Rendering of Height Fields, Technical report 96,
Georgia Institute of Technology, 1996
7. Meister M., H. Dan, Processing of Geographic
Data for CAAD-supported Analysis and Design
of Urban Areas, IGP- Swiss Federal Institute of
Technology, CH-8093 Zurich, Switzerland
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
N
f
2
8. (
C
L
9. C
S
C
A
10. F
F
14. R
C
E
12.5
C
g
13. 7