International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
remove borderlines of grassland. Furthermore, centerlines of 
wind erosion obstacles can be obtained by least squares image 
matching with roof-template (Zhang et al. 2003). 
  
Fig. 9. Grouped potential wind erosion obstacles 
3.4 Verifying with 3D information 
After image segmentation, line extraction and grouping on both 
images of one stereo, the matched lines are the centerlines of 
wind erosion obstacles. The conjugate lines on both images can 
be found with epipolar line and mean x-parallel of relatiwe 
orientation because there are only a few candidates of conjugate 
of the interested line. A global optimization is needed to make 
sure that no false corresponding exists. 
  
Fig. 10. Results of extracted wind erosion obstacles 
788 
The conjugate line pairs can be used to obtain the 3D lines in 
object space with known internal and external orientation 
parameters. Lines without conjugate on the other image can be 
initially projected onto a level plane with mean height of wind 
erosion obstacles. The new height value can be obtained from 
DSM with the projected plane coordinates. Then the image line 
can be projected onto a level plane with the new height. This 
recursive. search procedure usually converges within a few 
iterations. 
The obtained 3D lines are potential wind erosion obstacles. Roads, 
field boundaries, rivers and railways in GIS data are also potential 
search areas of wind erosion obstacles. All these information are 
compared with DSM to verify whether they are really wind 
erosion obstacles. At this point, the remained boundaries of 
grassland can be easily removed because grassland will be 
usually wider than wind erosion obstacles and thus a large area 
with same height information. Figure 10 shows the finally 
extracted wind erosion obstacles. 
4. EXPERIMENTS AND RESULTS 
In this section, experimental results of the proposed approach are 
presented. GIS data from ATKIS, stereo CIR imagery with known 
camera orientations, and DSM are used as sources of information. 
The general procedure of the proposed approach can be 
summarized as follows. Firstly, CIR image is classified into 
vegetation and non-vegetation areas, and non-vegetation areas are 
removed from the image. Afterwards, lines are extracted from the 
segmented image, followed by removing of extracted lines that 
belong to non-interested regions such as forests and buildings 
according to the available GIS-data. Then the remained lines are 
linked along-line-direction and then grouped cross-line-direction 
to get the centerline of wind erosion obstacles. Finally, camera 
parameters, matched lines, DSM and GIS data are integrated to 
derive true wind erosion obstacles. 
  
Fig. 11. Final results of extracted wind erosion obstacles 
P uere E 
er 
— 
Für f ft 089 M — © 
ms ng AN AA 
- ^um Re pres 11) 
r^ ov rnm 
KA