Geoscience Reference
In-Depth Information
Figure 3:
ditches to be typified (on the left, solid lines) and not to be typified (on the right,
dashed lines).
As we wrote in section two, not all the ditches of a dataset are part of a
group: some ditches run isolated, following the course of a road or sur-
rounding a field.
The first step of our algorithm is then to find which ditches belong to a pat-
tern and which do not: this is done analyzing the direction of each ditch
and then clustering them in groups. Depending on the way the data was
digitized, it can be hard to recognize a pattern; because of this the ditches
are preprocessed to ease the pattern recognition. During preprocessing,
every ditch is divided in segments with the same direction. Algorithms
performing the segmentation of lines already exist (e.g. (Plazanet 1995),
(Balboa 2009)) but the almost straight shape of ditches allowed us to set up
a quite simple algorithm that measures the angle between three consecu-
tive vertices and decides whether it is small enough to consider the three
vertices almost in-line, or otherwise to split the ditch in the middle vertex.
At the end of the preprocessing all ditches have been divided in almost
straight lines, that we will call segments; for each of these segments the
centroid and the average direction is computed. For two segments to be in
the same pattern they must have a similar average direction and their cen-
troid should not be too far away. The direction similarity and centroid dis-
tance threshold are controlled by two parameters of the algorithm: in our
tests for the generalization of 1:25000 scale data from 1:5000 scale data
we found respectively π /24 and 50m to be good values for them.
All the segments that are found to be part of the same pattern are then
grouped together in what we call a “ditch cluster”. At the end of the pro-
cess if a ditch cluster contains only one segment this segment will be
