International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 
at the shortest and one at the longest focal length depending on 
the available space around the object. Besides the cost of the 
digital camera, other data acquisition costs include the rent of 
the helicopter for one hour and surveying of 150 points (one 
day) with a Leica" total station. 
5.2 Modeling Results 
ShapeCapture" | [htp://www.shapecapture.com] software has 
been used for calibration, bundle adjustment, and image-based 
modeling. A general model was created from the aerial images, 
and several detailed models were created using sets of ground 
images, including models of the St. Martin's chapel, the towers, 
the loggia, the entrances, and details of the walls enclosing the 
courtyards. Figurcs 9-11 show snapshots of some models. 
  
  
  
    
Figure 10: Model of a castle entrance. 
  
Figure 11: Part of the model of St. Martin’s chapel. 
124 
5.3 Assessment of the Models Accuracy 
The castle structures extended over an area of about 100 meters 
by 64 meters and have up to 35 meters height. The castle is 
elevated by over 30 meters above the road approaching it. The 
aerial images were taken from an average range of 120 meters 
and average vertical distance of 65 meters above the castle. The 
field of view of each image usually covered at least all the 
buildings, with an average view width of about 160 meters. One 
pixel in the image corresponds to about 8 cm. The accuracy is 
assessed by the variance-covariance matrix of the adjusted 3-D 
points, and validated by surveyed checkpoints that had 3 mm 
accuracy. The achieved accuracy was in fact homogeneous and 
averaged: 17mm (X), 15mm (Y), and 16mm (Z). This is one 
part in 10,000 and represents 0.2 pixels. For the ground-level 
models, we achieved accuracies of: Imm to 2mm, which give 
average relative accuracy of one part in 6,000 and represents 
0.3 pixels. The superior relative accuracy reached from aerial 
images, even though they were taken from much longer ranges, 
is attributed to the better geometric configuration compared to 
ground images. Ground images had less than ideal locations due 
to obstructions and tight spaces around most buildings. 
5.4 Model Assembly 
Applying the procedure presented in section 4.5, we started by 
superimposing the general model on the floor plan (figure 12). 
Sections missing from the general model were added from the 
floor plan using known wall heights. Detailed models such as 
the gate in figure 10, and indoor models such as the chapel in 
figure 11, were then inserted. In figure 13, the gate, windows, 
and wall top edge connect the general model and the detailed 
model. Points along those common regions were first used to 
register the two models. Then only points along the top of the 
wall and bottom of the gate were kept in the general model, and 
new points were imported from the detailed model. The general 
model was then re-triangulated to account for the point changes 
and to create a hole where the detailed model was inserted. 
  
  
  
Figure 13: Assembling a detailed model into the general model. 
c 
   
Internation: 
lt ne 
Although tl 
we have re 
image-base 
from a low- 
with existi 
completely 
with realis 
assembly t 
20% of all 
80% were 
was much | 
use an accu 
higher leve 
advances to 
Acknowled 
We express 
Castelli Tre 
allowing us 
References 
Allen, P.K. 
historic sit 
23(6), Nov. 
Bacigalupo. 
reality for r 
Workshop 
Archaeolog 
Blais, F., 
developmetr 
Borg, C.E., 
approach tc 
Workshop « 
Corfu, Grec 
Cain, K., N 
for the Zav 
http://www. 
Cantzler, H 
by constraii 
University « 
DeLeon, V 
accessible r 
Conf. Virtu 
Scotland, S. 
Dikaiakou, 
Modeling 
Symposium 
Cultural He 
El-Hakim, : 
Photogrann 
pp. 443-448 
El-Hakim, : 
Proc. SPIE 
for 3D Sh: