may occur e.g. 
criteria are 
is used first to 
section points 
le best non- 
n-overlapping 
regardless of 
n is, that there 
icating a 3D 
‘ound. For the 
on points are 
points in that 
ines. The best 
extensions are 
ultiple images 
d to describe 
nages. Rather 
5, the task has 
ical walls and 
le in a few 
> nadir point. 
s, we believe, 
oof, which in 
ntense use of 
(s.a. there are 
plex (s.a. two 
ndicular lines 
les). There is 
nage features 
ct space, and 
by a simple 
intly. In spite 
ction, the 3D 
lained by the 
s shadow are 
or horizontal 
nain building 
1e horizontal 
s the vertical 
is however 
tions to be 
interpret for 
yuld require 
ample, it is 
'erpendicular 
or a plane to 
haracteristics 
n-conflicting 
lless of how 
is. This may 
at is used for 
evidence is 
ed earlier in 
buting to 3D 
line clusters by their contrast. It is also too restrictive, due 
to possible occlusion, to require that there should be 
contributing lines from all images to define a 3D 
horizontal line. Requiring contributions from less images 
would reduce the reliability of the lines and increase the 
risk to introduce false horizontal lines, but may still be 
necessary to include all relevant 3D lines. 
  
ep 
45790 + £0 ud 
45788 F 4 
45786 + = 
45784 | 4 
45782 F E 
45780 rF : 
45778 F zl 
  
  
À 1 
  
45776 i i i " 1 
79080 79082 79084 79086 79088 79090 79092 79094 
"ütep' m 
79095 
  
79080 
  
Figure 12. The selected non-conflicting plane hypothesis 
of the roof. Projected to XY plane (top), 3D view 
(middle), imposed on one image (bottom). 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
Although the list of cumbersome restrictions can be made 
long, there are reasonable solutions to them. The 
approach seems to be extendible without hitting dead- 
ends. The system will now be tested on other images. 
REFERENCES 
Axelsson P. (1994): Line Correspondence in Aerial 
Images Using Stochastic Relaxation and Minimum 
Description Length Criterion, ISPRS Commission III 
Symposium, Intl Archives of Photogrammetry and 
Remote Sensing, Vol 30, part 3, München. 
Braun C. (1994): Interpretation and Correction of Single 
Line Drawings for the Reconstruction of Objects in 
Space, ISPRS Commission III Symposium, IAPRS, Vol. 
30, part 3, München. 
Fuchs C., Fórstner W. (1995): Polymorphic Grouping 
for Image Segmentation, 5th ICCV, Boston, Mass. 
Fórstner W. (1994): A Framework for Low Level 
Feature Extraction, 4th ECCV, Stockholm, Sweden. 
Jamet O., Dissard 0., Airault S. (1995): Building 
Extraction from Stereo Pairs of Aerial Images: Accuracy 
and Productivity Constraints of a Topographic 
Production Line, In: Automatic Extraction of Man-Made 
Objects from Aerial and Space Imagery, Gruen A., et.al. 
(Edt' s), Birkháuser. 
Lin C., Huertas A., Nevatia R., (1995): Detection of 
Buildings from Monocular Images, In: Automatic 
Extraction of Man-Made Objects from Aerial and Space 
Imagery, Gruen A., et.al. (Edt' s), Birkháuser. 
McGlone J.C., Shufelt J.A. (1994): Projective and 
Object Space Geometry for Monocular Building 
Extraction, IEEE Conf. on CVPR, Seattle, Washington. 
Rissanen J. (1978): Modeling by Shortest Data 
Description, Automatica, vol. 14. 
Roux M., McKeown D.M., (1994): Feature Matching 
for Building Extraction from Multiple Views, IEEE Conf. 
on CVPR, Seattle, Washington. 
Wiman H., (1995): Object Related Image Segmentation 
using the Minimum Description Length Principle, Proc. 
of 9:th SCIA, Uppsala, Sweden. 
959