267 
In: Paparoditis N., Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds), 1APRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
multiple images and many lines are detected in images but none 
or very few are vertical or horizontal. Thus, we use the Lines 
Direction Ratio when the match ratio is close to 0. 
5.2 Wall verification results 
Combined memberships, for walls used for training and 
validation, have been computed as already described. The 
results in Figure 8 show that for actually existing walls the 
membership to class “wall exists” are close to 1 as expected 
while for demolished walls the result are close to 0, also as 
expected. There are some misclassified walls because occluded 
walls are included and will give a wrong result for single walls, 
but the overall merge for complete buildings is correct. 
5.3 Building verification results 
Using combined membership for walls the overall building 
membership has been computed as already described. The 
results Figure 10 show that for all existing buildings the 
membership to class “building exists” are close to 1 as expected 
while for demolished buildings the result are close to 0, also as 
expected. 
6. CONCLUSION AND OUTLOOK 
In this research it was shown that information on wall façades 
in oblique images can be used to reliably verify buildings in 2D 
vector dataset. Using the developed method, existing buildings 
result to memberships to class “building exists” close to 1 
while, for demolished ones, the memberships are close to 0. 
This method, besides providing the overall verification results, 
gives the result per wall, which may be used for updating the 
data if part of the building has been changed. 
Further work includes more experiments to improve the 
membership functions used for determination of existing and 
demolished walls. We are also considering the use of Support 
Vector Machines for robust combining of evidence. 
Our method requires a rough estimate of terrain and building 
height. For this purpose, a not very accurate DSM obtained by 
using the same oblique images (Gerke, 2009) may be used. 
Further work also includes incorporating evidence on roof and 
identification of walls occluded by other objects so that the 
search is limited to visible walls only. 
ACKNOWLEDGEMENTS 
We would like to thank BLOM Aerofilms for providing the 
images used. We also thank the anonymous reviewers for their 
comments. 
Lines Match Ratio memberships 
Lines Direction Ratio memberships 
Correlation Coefficient Ratio memberships 
о * — -г — 
0 0 1 0.2 0.3 0.4 0.5 0.6 
Correlation Coefficients - corners Ratio memberships 
0 0.1 0.2 0.3 0.4 0.5 0.6 
Figure 9: Fuzzy memberships - x-axis is measure and y-axis is 
the membership to class “wall exists” 
Figure 8: Wall memberships to class “wall exists” 
1 ‘ 
.9- 0.8 
ÜÜ 0.6 
f 0, 
Л 0.2 
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 
Building# 
♦ Demolished Training ■ Existing Training ▲ Demolished Validation • Existing Validation 
Figure 10: Memberships to class “building exists”