Hrabacek, Jan 
In table 1 the basic statistics of the models can be found. The summ 
there is the number of constraints specified b 
considered as one entity, 
Standard deviations used 
60 = 1°. 
ar 
ary contains the real number of constraints. In brackets 
y user, the difference is caused by the fact, that some constraints, however 
are rewriten as a set of several constraints (3 for the parallelograms, 2 for the plane parallelism). 
for weighting of the angle and the parallelism constraints are given in minutes (denoted ^), where 
  
  
   
Figure 4: Zürich data set — used images 
4.1 Models 
Model #1 
Figure 5 presents the result of the adjustment of the 
were no constraints employed, 
lacks regularity in the shape. B 
occlusions of upper parts of th 
one, could not be reconstructe 
of normal equations. 
geometrically weak data set, resulting in an irregular model. There 
apart from coplanarity constraints that force the points in their planes. The result obviously 
esides other facts presented in table 1, let us emphasise that only 4 images with significant 
e building were used. The dormer windows, that distinguish this model from the second 
d at all. The occlusions cause a lack of observations resulting in singularity of the system 
   
Figure 5: Model #1 — no constraints 
th 
to 
D 
ne 
R 
re 
ot 
€x 
va 
CO 
Model #2 
Figure 6 shows the result of computation with the same data set as used for the 
of geometrical object constraints. A summary of them is included in table 1 
constraints is equal .27, convergency was reached within 11 iterations. 
The constraint-based approach enables to obtain a complete model, des 
with poor redundancy. 
previous model, but with a strong support 
. The mean standardised residual of the 
A regularly shaped object model was obtained. 
pite occlusions and despite image measurements 
  
Figure 6: Model #2 — constraint support 
va 
In 
va 
PI 
va 
In 
4.2 Advantages of the use of weighted constraints 
4.2.1 Enhanced solution to occlussions 
The possibility to reconstruct occluded parts of an object using only line 
measurements has been described in (van de 
n Heuvel, 1999). It presents also an analysis of the minimum image line 
386 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000.