19 
re 
d 
1C 
an 
nd 
"n. 
he 
he 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
found, whereas in medium size cities the expected structure 
dominate, meaning that three or more major roads will meet in 
the city centre. In large cities the opposite trend can be 
observed: major streets will run around the city but in the centre 
itself only minor streets or even pedestrian areas will be 
located. 
This could be one useful information for setting up rules, which 
can be found automatically with data mining mechanism. 
: village or small town 
«= 5.000 — 10.000 residents 
  
small town 
ca. 40.000 residents 
  
large city 
ca. 520.000 inhabitants 
  
Figure 5. Typical arrangements of streets in the city centre 
depending on the dimension of the town. 
2. Junctions of roads have been investigated regarding the 
existence of nodes with four outgoing lines. The intention was 
to look into detail, if there are reasoning mechanisms to cut 
settlement areas into partitions, especially if the lines will meet 
approximately orthogonal (CRS type). 
Among other things it came up, that highways will be 
represented by separate clusters with solely one edge (ELL- 
junctions), with the exception of the access roads. Naturally 
there are only a few intersections with highways, the parts 
between are direct polylines without branches. 
As shown in figure 6 it could be one of a criteria to determine 
highways respectively to distinguish their access roads from 
their carriage-ways in data sets. It can be very helpful to 
validate further structures like the neighbourhood of settlement 
areas in the vicinity of an highway access. 
    
[= 
one edge cluster”, 
the highway is easy to locate in the middle. 
ec 
Figure 6. All red lines are 
339 
  
The analysis of junction or node types can also help to 
distinguish between different features on a geometric level: 
when looking at different linear networks, it gets clear, that 
certain junction types only occur with certain objects — or do 
not or only rarely occur with certain objects (figure 7). E.g. the 
4-junction mentioned above mainly can be found in road 
networks — and hardly ever in river networks, as in nature it is 
very rare, that four streams will meet in the same place. 
Another extreme example are lines which typically do not 
intersect at all or only at (very rare) saddle points. 
It does not lead to new knowledge, but to new information to 
the computer. This investigation can shed light on the content 
of a data set, especially which line elements belong to the road 
network. In this context the obvious rule can turn into a very 
helpful information. 
  
  
Figure 7. Appearance of different line elements: roads, rivers, 
administrative boundaries, contour lines 
Furthermore the investigation into the nodes with four outgoing 
lines led to following conclusion regarding a partitioning: these 
CRS nodes can be of a separating nature, especially along the 
major roads. It is similar to a Voronoi diagram, which here, 
however, does not exist on the basis of geometric distance, but 
rather on the topographic detail of the intersection of four lines. 
Figure 8 documents the results of the analysis in two different 
data sets, one French and one German data set. Especially along 
the major roads the data set is segmented into different 
partitions. In the figure on the left side you can see, that the data 
set is split in two main sections each on the left and on the right 
side of the picture. In the middle a valley with major roads and 
a town is located. 
dec 
  
Figure 8. Clustering of road networks by analyzing the CRS 
nodes. Left: French data set. Right: German data set. 
Other measures we are going to investigate is the "straightness" 
of a linear object, i.e. a collection of polylines that can be 
traversed more or less straightly. A method to derive these so- 
called strokes is described in Thomson and Richardson (1999) 
and has been used for network generalization and classification 
by Elias (2002). 
On the basis of described processes we are able to examine the 
data sets and the above mentioned results in more detail, 
whereby supervised and unsupervised models can hardly be 
kept apart at this stage. The following factors could be decisive 
for further (supervised and unsupervised) interpretations: size of 
a single mesh, length of segments between nodes, frequency of