Figure 4 - The statue surveyed 
5. THE SURVEY OF A SCULPTURE 
The above procedure has been applied to the 
determination of the digital surface model of a sculpture 
(see Fig. 4), as mentioned in the introduction. 
For the whole survey, 108 images have been acquired 
and 11 of the 150 targets used were determined by 
theodolite survey with an accuracy of 1/10 mm on all 
coordinates. 
The functioning of the proposed method has been tested 
with only 10 images, each containing on average 14 
targets, covering the sculpture's head. 
Control points were available only in the front part of the 
body. The 10 images were successfully tied together. All 
targets were automatically localized and measured on 
the images, while the control points were labeled by the 
operator (see Table 5). 
Looking in all possible sets of three images, thanks to 
the epipolar constraints, the coordinates of additional tie 
points were determined (see Table 6). 
  
  
  
  
image n. 10 
targets 9 
numbered 3 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Table 6 - Results of the target location procedure for 
group 1 (in dark cells) 
Each of the four remaining images has been linked to 
one of the three sets of group 1, by manual identification 
of the minimum number of common targets. 
Correspondencies between the targets on the new 
images and the images of group 1, were automatically 
found (see Table 7). Including these new observations, a 
new bundle adjustment have been run, improving the 
exterior orientation. 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
image n. 10 
targets 9 
numbered 8 
  
  
  
image n. 1 2 3 4 5 6 7 8 9 | 10 
  
targets 171.16 | 14 | 14 | 14-12} 15 | 16 | 16 | 9 
  
  
  
  
  
  
  
  
  
  
  
  
  
c. points 6 T 6 6 4 4 | 6 6 4 | 3 
  
Table 5 - Number of targets and control points 
in the test images. 
Three pairs of images (group 1) were selected out of the 
whole group, which contained at least 3 c.p. and 3 tie 
points; a bundle adjustment has been computed, 
providing the approximate exterior orientation. 
Table 7 -Results of target location on the whole set 
6. FINAL REMARKS AND PERSPECTIVES 
The results shown above confirm that the implemented 
procedure is effective in speeding up the measurement 
of tie points for image orientation. Many developments 
can be still introduced either robustifying the algorithms 
or improving the user-interface. 
The search for the approximate orientation may be 
simplified and made capable of dealing with strong 
convergent cases. 
At present the procedure runs over all possible sets of 
three images to search for additional target 
correspondencies. This is still acceptable with a small 
number of images, but the combinations grow 
exponentially: therefore we need to discard all sets 
whose images cannot share, on the basis of the exterior 
orientation, any common target. 
Aknowledgements 
The authors are grateful to Carl Zeiss GmbH and to 
Dipl.-Ing. Torsten Kludas for making available the 
images used in this work. Many thanks go also to Dr. 
Sergio Grassi for his support in program developments 
under X-Windows. 
References: 
Ayache, N., 1991. Artificial vision for Mobile Robots. MIT 
Press, Boston, U.S.A.. 
Baltsavias, E.P., 1991. Geometrically Constrained 
Multiphoto Matching. Mitteilungen N. 49, Institute of 
Geodesy and Photogrammetry, ETH, Zurich, 
Switzerland. 
522 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996 
Balts: 
1990. 
Photc 
Photc 
Fórst 
algor 
Finlai 
Fôrst 
Imag 
Weel 
Gern 
Klude 
recor 
Meas 
Jena 
Lemr 
Tech 
B8/III