Annett Faber 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
   
  
  
  
  
  
  
  
  
  
  
  
   
  
  
  
  
  
  
  
  
(a) City map of New York (b) MOMS-02 image of Ballarat 
(Australia) 
he 
he Figure 5: Example data sets 
).5 
j | 
zm vut. 
b SE 
we L 
f | 
0 : 
in Fr à I; 
WO 
(a) Results of line detection (b) Regions of robust estima- (c) Segmentation 
tion 
Figure 6: Results for the city map of New York 
ce | 
rat | 
f / (a) Results of line detection (b) Regions of robust estima- (c) Segmentation 
9 tion 
are Figure 7: Results for the MOMS-02 image 
her 
15 
6 CONCLUSION 
One may proceed in two directions to improve the results of the algorithm: 
NS e Regarding to the results described above, as a first step we have to investigate into the automation the choice of an 
| adaptive size of the window depending of the density of the road network. 
ole 
lly e Replace the algorithm for transferring the orientation data from the roads to the building blocks by a distance trans- 
es. form carrying the orientations into the neighborhood. This will significantly increase the robustness of the procedure 
the with respect to variations in density. 
gi International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 279