965 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
during the field campaign, using a total station, were used to 
define the scale and the common coordinate system for a joint 
representation of all datasets in the final model. The standard 
deviation of the bundle adjustment was 1.9 micron, which is 
less than half a pixel in image space. The expected accuracy in 
object space was approximately three centimetres. Using this 
data, a 3D model of the empty niche could be generated. 
The dataset A had to fulfil two mayor aims. It provided 
information for the camera calibration and the exterior 
orientation of the images of the dataset B and secondly, it 
enabled us to visualize the reconstructed statue in combination 
with the niche and its immediate surroundings. 
4.2 Dataset B 
4.2.1 Calibration and orientation: The two analogue 
middle format images were scanned with a resolution of 10 
micron, using the photogrammetric scanner Vexcel UltraScan 
5000. The missing interior orientation parameters had to be 
determined by a simultaneous calibration, under consideration 
of the limited information provided by the two images. 
Therefore, the dataset B was combined with the dataset A by 
manual measurement of tie points. An automatic tie point 
measurement was conducted but the result was not usable, 
because of too large perspective differences between the two 
datasets. Figure 4 shows the distribution of the tie points 
between the two datasets. 
Figure 4. Tie point distribution between dataset A (upper row) 
and B (lower row) 
In order to stabilize the system further, tie points between the 
images of the dataset B were measured using two different 
procedures. First, a manual measurement followed by the 
automatic tie point generation was performed, both conducted 
with the photogrammetric software LPS (Leica 
Photogrammetry Suite). The automatic tie point measurement 
worked in an acceptable way. Nevertheless, under 
consideration of the poor network, using only two images, a 
second procedure was applied to increase the number of tie 
points. The SIFT operator (Lowe, 2004) was applied to find 
additional points. To improve the quality and number of 
measured points, the images were enhanced by using an 
adaptive histogram equalisation. Afterwards the SIFT operator 
was applied in a patch-wise mode to reduce processing time, 
memory consumption and the risk of mismatches. This 
procedure generated approximately 1700 points distributed over 
the whole image. Using the above mentioned software SGAP, 
a combined bundle adjustment of the datasets A and B was 
conducted. To preserve the high accuracy of the first dataset, 
the interior as well as the exterior orientation parameters of the 
dataset A were kept fixed. Consequently, the dataset B was fit 
into A using a constrained bundle adjustment. After an iterative 
procedure to eliminate mismatches of the SIFT procedure, a 
sigmaO of 0.9 pixel was obtained for the whole system. The 
accuracy of the exterior and interior orientation parameters, as 
well as the accuracy of the determined lens distortion 
parameters (Brown model) was sufficient to go ahead with the 
next processing steps. The lens distortion, as well as the shift of 
the principle point were removed from the images to avoid a 
wrong interpretation of the distortion parameters in the 
subsequently used software packages. 
4.2.2 Automatic Measurements: The automatic Surface 
Model (DSM) generation was conducted using the software 
SAT-PP (SATellite image Precision Processing) (Zhang, 2005, 
Remondino et al., 2008). The results are shown in Figure 5. 
Because of the limited image quality, the small number of 
images (two), the oblique viewing directions and the small 
base-distance ration, the measurements were not usable to 
model the statue with sufficient accuracy and resolution. 
Nevertheless, some parts of the data, especially the breast and 
leg region, could be used for some following processing steps, 
e.g. the registration and determination of the equidistance of the 
contour line map. 
Figure 5. Result of the automatic image matching