ul 2004 
with the 
rigidity 
oundary 
em: The 
he snake 
different 
he snake 
i virtual 
a hilly 
utialized 
/ drags it 
1odel, an 
ates in a 
se of an 
d, is the 
y Bx ds 
cribed in 
the here 
(13 
to select 
Lo control 
irc length 
to lead to 
d as well 
its. of the 
nally, the 
ntroduced 
rm "deep 
aries (cf. 
vided into 
of buffers 
allows a 
values can 
of dense 
her DSM- 
ial wind 
extracted 
yurse scale 
is carried 
's than the 
f the wind 
+ extracted 
minimum 
is the use 
| jn a finer 
in (Straub 
ts to lines, 
International Archives of the Photogrammetry, Remote 
4. RESULTS 
In this section example results from an application area are 
presented. In. Figure 4 the red channel of the imagery of a 
selected region of interest is depicted, the boundaries of the 
regions have been derived from the ATKIS-data. The 
preliminary results of the segmentation are shown in Figure 5, 
cach grey value is one segmented field area. The field arcas arc 
superimposed with extracted lines in white and additional 
further knowledge from the GIS-data in black. The final result 
of the segmentation step is depicted in Figure 6. It must be 
recognized, that the geometric correctness is not always 
satisfying: For example, the field boundaries in the middle top 
of the depicted region of interest in Figure 6 are not exactly 
identical to the real boundaries of the field in the image. The 
initialized snake algorithm is a reasonable possibility to refine 
these preliminary results, details for one field and the associated 
iteration steps of the snake algorithm are highlighted in 
  
Figure 4. Selected region of interest, depicted is the red channel 
of the acrial image 
  
Figure 5. result, 
Preliminary 
Superimposed with extracted lines (white) and 
further knowledge (black) 
segmentation additionally 
1069 
Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
Figure 3. The refined result of the extraction of field boundaries 
is depicted in Figure 7. The geometric correctness has 
improved, but the topological correctness has in some parts 
suffered. 
The realization of the proposed strategy extracting wind erosion 
obstacles is still in progress, which is why results in this domain 
are not depicted. 
5. CONCLUSIONS 
In this paper work on the automatic extraction of field 
boundaries and wind erosion obstacles from aerial imagery is 
presented. A semantic model integrating the different objects to 
be extracted and the GIS-data is described. Next, the derived 
strategy is discussed considering an automatic processing flow: 
The strategy is divided to extract the objects of interest 
  
Figure 6. Result of the preliminary field boundaries is depicted 
in black 
  
Figure 7. Refined result of the field boundaries after the snake- 
processing: Initialization is depicted in white, field 
boundaries are depicted in black