ation. À 
  
  
  
  
  
  
  
  
  
1e paper 
1. 
exterior 
odels. 
odels for 
ts which 
in most 
ant areal 
nes. Be- 
t. Edges 
mes the 
iy mark- 
weather. 
of build- 
the 3D- 
n. Using 
s. Other 
Figure 1: An example of the pose clustering 
| for au- 
up of a 
re of the houses of the old database, additional buildings or groups of The redundancy of the spatial resection is used to de- 
|| system houses are chosen for a higher redundancy and robustness of tect and correct a false position of control point mod- 
artment) the resection. els. 
A robust spatial resection fits all the models in an op- 
oto base 3 timal way using the correspondence of the 3D model 
THE ORIENTATION PROCESS ; 
tha ew edges to the image edges. 
e scale of The module for the automatic exterior orientation (AMOR) Finally a selfdiagnosis verifies the result with respect to 
program was developed by Wolfgang Schickler and consists of the precision and sensitivity. This enables the automatic 
lished to following steps [Schickler, 1994]: procedure to decide whether the determined orienta- 
resection tion parameters are acceptable or should be rejected. 
t control 
building. * With the help of approximate orientation values the Tests with AMOR, using the already measured wireframes, 
the more 3D-edges are projected leading to approximate posi- show different criteria which influence the success of the ori- 
a spatial tions of the 2D-model edges in the image. entation. The following list gives an overview. 
* An edge extraction is computed for the parts of the e The most important criterion is the size of the search 
lance the image where the control points should be searched. area for the models or, in other words, the accuracy 
d * A pose clustering is used to make an approximate ee nate als givenio AME. 
st IESU TO search for the-2D-position of the control point mod- e Ce Puta on time fort ee ge A anche 
ES RA els in the image. The model edges are compared to oes c N EA han FIER icant yw en ors N AUS 
S ES the image edges. In Figure 1 one can see an exam- idi se ; x En Sine ma en e A : 
ection ple of an image part with image edges and the model s ity'o m Ing 3 contre pom mo e 1 ecomes mgner 
Institute which should be found. The result of this process is ecause ere ore more non-buiding edges ; 
sides tire the position where the number and the quality of the The probability of the success of AMOR could be in- 
correspondences reaches its maximum. At least four 
control point models must be found. 
219 
creased by increasing the number of control point mod- 
els. We use on average 9 models per aerial image. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996