AUTOMATIC
V. Walter, S. Volz
Volker.Walter@ifp.uni-stuttgart.de
Clustering
spatial data can be interpreted as segmentation or classification processes
which find meaningful patterns within geospatial databases. Such patterns can
be used to adapt further algorithms to the individual characteristics of the
detected classes or clusters. For example, cartographic generalisation works
differently in urban and in rural areas. If this kind of information is not
explicitly available in the database, it can be derived automatically with
clustering algorithms.
The area of
spatial data clustering has been extensively studied and various approaches are
available. However, to the best of our knowledge, none of the existing
techniques has tried to perform the clustering of vector data in the raster
world. As we will show in this paper, this is a simple and straightforward
approach that allows a fast computing of clusters.
In our
study, we use vector street data from the Geographic Data Files (GDF), in order
to derive clusters of different degrees of urbanity. At the beginning of the
process, an operator can define two different parameters for generating the
clusters: the grid size of the resulting raster map and the radius around the
centre of each grid cell (cluster
radius) so that the area for which the cluster indicators have to be
observed can be calculated (area of
influence). As indicators for recognizing different levels of urbanity,
we use node density and rectangularity of streets, since we
assume that (at least in Germany) in city centres there are usually more
topological nodes and more irregular, non-orthogonal streets than in suburbs or
rural areas.
After the
operator has chosen the clustering parameters, the whole area of investigation
is subdivided into equally sized, square-shaped grid cells. Then the area of
influence is determined and the indicators, i.e. node density and
rectangularity, are calculated. The result is a raster layer for each
indicator. In order to join the different layers and to achieve a final
categorization of each individual grid cell, a function has to be defined
enabling the combination of the different raster layers. Before joining the
different raster layers, it is possible to pre-process them with image
processing techniques. For example a Gaussian filter can be used to smooth the
raster layers or to decrease noise. This can also be done with the final result
raster layer.