Cl PA 2003 XIX th International Symposium, 30 September - 04 October, 2003, Antalya, Turkey 
Figure 5: 3D point cloud generated with VirtuoZo automatic 
matching on the metric images (ca 178 000 points). 
3.2.2 Automated measurements with our software 
A multi-photo geometrically constrained (MPGC) least squares 
matching software package, developed at our Institute, was 
applied to the metric images [Gruen et al., 2001, 2003]. The 
automatic point measurement works according to the following 
procedure: 
1. Selection of one image as the master image. In our 
application, the center image was selected; 
2. Extraction of a very dense pattern of feature points in the 
master image using the Foerstner operator; 
3. Cross-correlation for each feature point to get the approxi 
mate matches for the following matching procedure (using 
also the epipolar geometry determined by phototriangula 
tion); 
4. MPGC matching for fine measurement, including patch re 
shaping parameters. MPGC exploits a priori known geomet 
ric information on orientation to constrain the solution and 
allows for the simultaneous use more than two images 
[Gruen, Baltsavias, 1988; Baltsavias, 1991]. 
In our application, for each feature point in the master image, 
all 3 metric images were employed for matching. With the 
MPGC approach, we can get sub-pixel accuracy matching re 
sults and 3D object coordinates simultaneously (Figure 6, left) 
and also, through covariance matrix computations, a good basis 
for quality control. 
Figure 6: The GUI of our MPGC matching software, with the 
matching results and the computed 3D object coordinates (left). 
The measured point cloud (right). 
The procedure resulted in fairly reliable and precise matching 
results. 49 333 points (without the surrounding rocks) and 73 
640 points (with part of the surrounding rocks) were obtained. 
The point cloud data is shown in Figure 6, right. Although we 
use an automatic blunder and occlusion detection, some blun 
ders are present in the final 3D point cloud. Moreover, there are 
some gaps in the cloud, mainly due to the shading effects 
caused by the variation of the illumination conditions during 
the image acquisition. Furthermore, many folds of the dress 
could not be reconstructed automatically, therefore these impor 
tant small features had to be measured manually. 
3.2.3 Manual Measurements 
The dress of the Buddha is rich in folds, which are between 5 
and 15 cm in width (Figure 7). The automated procedures could 
not recover these small details, therefore only precise manual 
measurements can reconstruct the exact shape and curvature of 
the dress. 
Figure 7: A closer view on the folds of the dress of the Buddha 
(left) and how they were constructed (right). 
We imported the metric images in the VirtuoZo stereo digitize 
module [Virtuozo NT, 1999] and performed manual 
stereoscopic measurements. Three stereo-models are set up and 
points are measured along horizontal profiles of 20 cm 
increment while the folds and the main edges are measured as 
breaklines. With the manual measurement a point cloud of ca 
76 000 points is obtained and the folds on the dress are now 
well visible (Figure 8). 
Figure 8: The point cloud of the manual measurement. 
The main edges and the structures of the folds, measured as 
breaklines, are well visible. 
3.3 The modeling process 
3.3.1 Automatic measurements 
Due to the smoothness constraints and grid-point based 
matching, in both automated procedures the small folds on the 
body of the Buddha are not correctly reconstructed and the