International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
are of great importance in the context of thresholds for syntactic
pattern recognition. There are several regulations which
determine thresholds for minimum widths or distances between
objects (Hake et al., 2002).
In digital cartography, the type of representation is not
restricted to the output of printed maps. Furthermore, electronic
screen displays have to be taken into consideration. In
comparison to printed maps, the electronic visualization is
limited by the size of the usable area on screen as well as the
smaller resolution (Brunner, 2001). The interaction of input and
output media has to be taken into account with regards to the
resolution and the graphical minimum size has to be determined
accordingly.
The preceding considerations point out, that important
topographic objects have to be emphasized during
generalization to ensure their perceptibility in different scales.
In a three-dimensional model this is all the more the case, as
objects are easily hidden by a surface representation in the third
dimension. After emphasizing certain objects within the model,
other objects have to be displaced. The remainder of the article
is structured as follows: firstly a survey of significant research
work is given, secondly the applied approach for the
generalization process is presented. Finally, the article closes
with a conclusion and suggestions for future work.
2. RELATED WORK
As advances in technology give rise to further growing sizes of
data sets, automatic simplification techniques for highly
detailed surface models are of considerable interest. Over the
past years, several effective techniques have been developed,
providing powerful tools for tailoring large datasets to the needs
of individual applications and for producing more economical
surface models (Garland, 1999). The often cited survey from
Cignoni et al. (1998) compares different mesh simplification
algorithms and gives a good overview on existing methods.
Heckbert & Garland (1997) also present a comprehensive
summary on polygonal surface simplification, in which they
attempt to categorize previously described algorithms. Luebke
(2001) describes and evaluates the most important
simplification algorithms from a developers point of view.
Automated cartographic generalization with the aim of
consistency and reproducibility of the generalization operation
is among the most challenging tasks in digital cartography. For
a fairly long time, numerous research work has been concerned
with this problem (Buttenfield & McMaster, 1991; Muller et al.
1995). A large amount of research work has especially focused
on the generalization of two-dimensional databases (Beard,
1991; Powitz, 1993; Ware & Jones, 1998). The generalization
of linear objects is mainly investigated for street objects and
river networks (Rusak Mazur & Castner, 1990; Thompson &
Richardson, 1995; Jiang & Claramunt, 2002). In this context,
selection and simplification of objects are the most frequently
discussed issues (Douglas & Peucker, 1973; McMaster, 1987;
de Berg et al., 1995; Neyer, 1999).
Another basic operation within the generalization process is the
enhancement, which is the enlargement or widening of objects.
These operations mainly have the purpose of maintaining
legibility, whereas there are scale dependent minimum sizes of
lengths and areas to consider (Hake et al., 2002). In the course
404
of enhancement, the geometric correctness of the objects is
neglected for the benefit of the enlargement out of scale.
Moreau et al. (1997) present a generic method for the
broadening of streets, river networks and other linear objects
around their middle axes. This method uses the semantics of the
object definition in addition to several automatic rules for the
adjustment of the geometries of neighbouring objects. Besides
the middle axes, cross sections are implicated during the
broadening process. A street can be approximated using a
sequence of segments and the relevant widths for this segment
characterized by a profile.
In connection with enhancement operations, displacement is a
further fundamental generalization process. One of the
approaches under examination is based on last square
adjustment (Harrie, 1999; Sester, 2001). Others concentrate on
finite element methods (Bobrich, 1996; Hojholt, 2000).
Additionally research is undertaken in the field of object
displacement in raster data sets (Jäger, 1990; Li & Su, 1997).
Generalization operations for three dimensional data sets is
primarily investigated in the context of the generalization of
buildings (Sester & Klein, 1999; Lal & Meng, 2001; Thiemann,
2002). Furthermore, several research is concerned with relief
generalization (Wu, 1981; Weibel, 1989). On the one hand they
work with contour lines, on the other hand raster data sets are
used.
One possibility for the generalization of surfaces in the raster
format is filtering. Low pass filters fill the gaps between
homogenous regions. This leads to a fusion of objects,
simplifying the existing representation (Bartelme, 2000). For
the generalization of raster data, several morphological
operators for the widening or reduction of objects exist (Li,
1994). During these procedures, the original pattern is shifted in
all directions given by the metric of the operators, which for
instance, results in a smoothing of lines.
3. GENERALIZATION OF THE DTM
To maintain the perceptibility of street objects in the DTM
during the generalization process, they have to be emphasized.
These procedure and the resulting problems are discussed in the
following. Streets are emphasized in such a manner, that they
are still perceivable from a given distance. Thus the distance to
choose depends on the “scale”.
3.1 Data Sources
3.1.1 ATKIS DLM: With the Authoritative Topographic
Cartographic Information System called ATKIS, the Federal
Republik of Germany holds a nation-wide database for two and
three dimensional spatial data (AdV, 1998). One of the products
within the framework of ATKIS is the Digital Landscape Model
(ATKIS-DLM), which groups topographic objects in logical
units with regard to the feature type. The geometric properties
of the ATKIS objects such as position, size and shape are
expressed through two-dimensional vector based descriptions
using primitives like points, lines or faces. Factual data are held
within the attributes and relations between objects respectively
object parts. The vector data representing the streets is taken
from the base-DLM with a scale of 1:25 000, the planimetric
accuracy is stated to be better than 3 m.
Internati
err
In the fe
catalogut
road traf
lanes b:
function,
characte
3.1.2
stereo Pl
for gei
(Knaben
refer to ]
the obse
Model (1
well as [
or build
points, a
The poir
meter.
1999b; |
with 0.1.
The dat:
part of tl
3.2 En
Based o
data set,
axes. Th
raster ce
the heig
axes. In
of the ei
Stuttgart
triangles
10 000.
streets a
the ima;
percepti
perceiva
represen
Altering
consider
between
visible i
broaden
appear 1
The firs
of adju
interpol:
an optic
the su
interpol:
interpol.
between
value in
approac
the bro.
simply
interpol.
Streets ¢
took p
enhance
overvie
