The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi XXXVII. Part B5. Beijing 2008 
966 
4.2.2 Manual measurements: The manual measurements 
were conducted with the above mentioned software LPS. The 
aim was to preserve especially the folds of the dress and other 
small, but important items. Because of the local topography the 
images were acquired with very oblique viewing direction and a 
small base-distance ratio. This led to occluded areas in the 
upper parts of the shoulders, as well as at the upper parts of the 
arms, see Figure 6. The lower legs and feet were not visible at 
all. Therefore, the image-derived 3D data was not sufficient to 
model the whole statue without any other information. 
Figure 6. Manual measurements with marked occlusions 
4.3 Dataset C - contour map 
To solve the problem of occluded parts, the third dataset - a 
contour map published in (Higuchi, 2001) had to be used. 
Describing a truly 3D object by a 2.5D contour map will cause 
some problems. In steep regions, the contour lines merge, 
which makes it difficult to interpret them correctly. In addition, 
information about the equidistance was missing. The latter 
problem could be solved using manual measurements. The 
difference in height between two areas, visible in both datasets, 
was measured from images and compared to the number of 
contour lines. Thus the equidistance was estimated with 10 cm. 
Using this information the contour lines were digitised by 
scanning and applying an automated line following routine 
from ArcMap. The absolute depth values were introduced using 
a starting point with an approximate depth, defined using the 
manual measurements. The derived point cloud covers the area 
from the middle of the breast down to the knees, which is 
jointly covered by the image-derived data. Additionally, the 
occluded shoulders, the head and some other small parts of the 
statue are covered. 
5. COMBINATION OF THE DATASETS 
The co-registration of the datasets A and B was already done 
during the joint bundle adjustment. Both datasets are present in 
the same coordinate system. 
The co-registration of the dataset B and C was conducted in the 
following manner. The contour lines as well as the manual 
measurements were converted into dense 3D vectors with a 
point interval of one centimetre, to preserve the high frequency 
information. These two structured point clouds were co 
registered using the software Geomagic Studio in a two step 
procedure. First, initial transformation parameters were 
calculated by manual measurements of common points. In a 
second step the two surfaces were accurately co-registered by 
the ICP procedure of Geomagic Studio, see Figure 7. After co 
registration the mean distance between the two datasets was 9.9 
cm. 
Figure 7. Co-registration of surfaces using Geomagic Studio 
a) Manual image measurements 
b) Digitized contour lines 
c) Co-registered datasets 
The difference of around 10 cm in the registration result was 
mainly due to the folds, which were included in the manual 
measurements, but not in the contour data. Concerning the fact, 
that the depth of these structures is between 10 cm to 20 cm 
above the average surface depth and the manually measured