Figure 4.4. Gaussian feature 
4.5. Hyperboloparaboloidal features. Terrain morphology 
modelled as hyperboloparaboloidal surfaces can only be 
modelled via selective modelling when Az/z > 6.0%. 
Applying the optimum modelling, the accuracy was 
improved by 0.13% to 1.33%, and the efficiency by 6% to 
19%. 
  
4.5. Hyperboloparaboloidal feature 
4.6. Ridge line features. Terrain morphology modelled as 
ridge line can only be modelled via selective modelling 
when Az/z > 2.0%. Applying the optimum modelling the 
accuracy was improved by 1.6% to 3.6%, and the efficiency 
by 10% to 27%. 
  
Figure 4.6. Ridge line feature 
4.7. Fauit features. Terrain morphology modelled as fault 
surfaces can only be modelled via selective modelling when 
Az/z : 2.0%. Applying the optimum modelling, the accuracy 
was improved up to 9.4%, and the efficiency up to 25%. 
4.8. Combposite features. Terrain morphology modelled 
as a combination of those surfaces can only be modelled 
via selective modelling, when Az/z - 5.0%. Applying the 
optimum modelling the accuracy was improved by 0.1% to 
0.28%, and the efficiency by 4% to 13%. 
796 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
  
    
  
  
  
  
  
   
   
  
  
  
  
  
  
   
    
  
   
      
  
  
  
   
     
    
   
    
   
   
   
   
    
   
    
   
   
   
  
Figure 4.8. Composite feature 
5. OPTIMUM SAMPLING APPLIED TO REAL 
TERRAIN RELIEF 
To verify and consolidate the conclusions drawn from the 
experiments using artificial ideal geometric primitives and 
their composites some experiments using real terrain 
morphology were conducted in the Bonnieux region (south 
of France). 
This region is partly covered by flat and partly by accidental 
terrain. This justifies the use of optimum morphologic 
modelling. The Easting of the area was between 840 200 
and 841 800, the Northing between 174 000 and 176 880, 
the altitude of terrain between 482.000 m and 243.000 m. 
The terrain relief was represented by 16384 points. Two 
areas with some abrupt changes have been delimited from 
a more homogeneous terrain, and Z information was 
collected selectively using the MAPS 200 system. This 
information contained 382 points in vector form. 
  
  
  
variant o] MAXER Pts 
[1 696 of z 2.78 0.65 
opt 8% of z 5.87 0.71 
  
  
  
  
  
  
Table 1 performance estimates for optimum versus semi 
automatic modelling 
  
Test R R R 
O max E 
  
  
  
  
  
  
opt tag OCT 
  
Table 2: Performance estimation of different variants of opt 
with respect to IN 
In conclusion the following can be stateed: the fidelity of 
the representation is improved by inclusion of X information. 
Apart from a great improvement in the accuracy of the 
Skeleton information, the overall accuracy and overall 
efficiency are also improved significantly, compared to semi 
-automated modelling ( R 0= 33% and R E = 10% ). Finally, 
for this region, we observe that by including the I 
information which fulfils the specifications of the rule base 
we can get, not only a better modelling, but also higher 
accuracy, with less effort. 
6. CONCLUSIONS 
to] 
eil 
ati 
in 
CO 
thi 
sl 
thi 
fo 
as 
RI 
Ay 
cl: