itofa 
> may 
ption, 
f the 
riori 
tation 
tional 
out a 
le are 
of the 
of the 
; from 
era or 
1 how 
ject's 
en 3D 
2d, is 
ketch 
ledge 
pecial 
ledge, 
m are 
|a 3D 
to the 
d 2-D 
ich of 
nding 
y the 
lanes 
E are 
. The 
aints, 
n the 
on the 
anual 
cases 
ution. 
  
  
  
  
  
  
  
  
  
  
  
  
; A 
vi 
Iput | > Intermediate data — | — — > | Output: 
2D sketch + component of declaration 3D model 
~7 V. 
* 
\ 
. 
N 
2 
Knowledge base 
27 
3 
4 
ve 
o 
d : 
^ » x 
^ E 
^ x 
5 
«------. > 
- 
* ^ 
s ^ 
N c 
  
Control modul 
  
  
  
Fig. 6 Components of the rule-based system 
As there is no apriori knowledge about the object at the 
beginning except the incidence relations or optionally about 
the camera or exterior orientation, the reasoning can only be 
started by building up hypothesis about geometrical 
relationships. In case of redundant information the results can 
be improved by calculating the values for the attributes in a 
simultanous evaluation process. If an operator is asked to 
indicate geometrical constraints, the system might have more 
information available as necessary. 
The following strategy is implemented as one part of the 
control modul, solving the 2nd task of the reasoning process 
(cf. chap. 3.1): 
Initially E contains all unknown planes. 
repeat 
repeat ( *internal reasoning* ) 
for all planes in list E do 
if plane determinable then perform 
* plane determination 
* point in plane determination 
*line in plane determination 
* delete plane from E 
endif 
endfor 
until no plane in E is determinable any more or E is empty 
if E is not empty then (* operator request *) 
ask the operator for a spatial relation 
endif 
until all planes of the object are determined in space 
There is freedom in the sequence of planes in line 4 of the 
algorithm which may be used to optimize the interpretation 
with respect to stability. The operator request may be assisted 
S19 
by the system, providing a list of candidate hypothesis from 
which the operator has to choose one. This may again be used 
for optimization. Finally all not yet used hypothesis may be 
combined to determine best estimates for the spatial 
coordinates of the object. 
4.4 Examples 
The reasoning for a collection of houses is discussed using the 
example in Fig. 7. The bold lines are classified as lines whose 
direction cosines in space can automatically be derived by 
locating vanishing points (Fig. 7a). These are the three 
vanishing points pointing downwards, backwards and to the 
left. The system now calculates the parameters of all those 
planes which fullfill the criteria for determination. These are all 
planes with at least two bold lines not being parallel (cf. rule 7 
in chap. 3.2). As during the first pass the direction cosines of 
the undetermined lines in space (thin lines) could not be 
calculated, the parameters of the remaining planes were 
determined by the system in a second pass, still within the 
internal reasoning (cf. control modul, line 3). Therefore in this 
example it is not necessary for an operator to indicate any 
geometrical relations. The result of the collection of buildings, 
generated by an alternative view, is shown in Fig. 7b. 
  
  
  
  
b. 
Fig. 7 A collection of houses 
a. Perspective line drawing 
b. Representation of the reconstructed collection of houses