CIP A 2003 XIX th International Symposium, 30 September - 04 October, 2003, Antalya, Turkey 
Within a project supported by the Swiss National Science 
Foundation and local cultural authorities, the relief and its 
constructor Franz Ludwig Pfyffer are subjects of an inter 
disciplinary research. The Institute of Geodesy and Photo- 
grammetry of ETH Zurich is involved with following tasks: 
• 3D reconstruction of the relief for the documentation of 
the cultural heritage. 
• Analysis of the relief accuracy for the purpose of research 
in the history of cartography - a comparison of the 
reconstructed virtual relief with current map information 
(digital maps and terrain models). This work also 
includes analysis of the old maps of the region which 
were created at the end of 18 th century. 
• Investigation on Pfyffer’s surveying and model 
constructing procedures. 
3. 3D RECONSTRUCTION AND VISUALIZATION 
The project objectives require a high quality 3D model of 
Pfyffer’s relief in terms of accuracy. The overall height 
difference between the reconstructed model and the original 
may amount to a maximum of 1 mm. Represented by around 
11.5 m in reality, this value approximately corresponds to the 
accuracy of available national DTM in alpine regions. Thus, a 
comparison of the historical and present terrain models in the 
analytical project phase (see Chapter 4) can bring reasonable 
results. 
3.1 Data acquisition 
The relief is situated in a narrow cellar room without natural 
light on a table about 65 cm over the ground. These conditions 
make the image acquisition as well as control point 
measurement rather difficult. From a construction platform 
fixed at the height of 185 cm above the relief surface, 87 
analogue and 50 digital “aerial” coloured images were taken, 
both providing stereo overlap for the whole object (Figures 2 
and 3, Table 1). The reason for such a double-block image 
acquisition was to make sure that the image quality will match 
the expectations because no new acquisition would have been 
possible any more. Both cameras were calibrated using a close- 
range photogrammetric reference field. To avoid disturbing 
specular reflection of the shiny relief surface, spotlights in a 
combination with dispersion umbrellas were used for indirect 
illumination. In spite of this, particularly the digital images 
show up rather big reflection and worse radiometric quality 
than their sensitive film-based counterparts. 
Figure 2. Image acquisition 
The three-dimensional position of 40 marked control points 
temporarily placed onto the relief surface was determined by 
theodolite measurements in a local coordinate system with an 
accuracy of 0.09 mm. Considering the unfavourable space 
conditions leading to flat and wide-angled rays this result is 
satisfactory and sufficient for further processing. 
Figure 3. A part of an analogue image (scanned with 1270 dpi) 
Camera 
Analogue 
Rollei 6006 
Digital 
DSC460c 
Calibrated focal 
length 
83.557 mm 
28.871 mm 
Number of images 
87 
50 
Number of strips / 
number of images per 
strip 
4/13 
+35 additional 
images in 
mountainous 
areas 
5/10 
Image size 
6x6 cm 2 
3060 x 2036 pixel 
Scale / Pixel footprint 
1:23 
0.6 mm 
Table 1. Parameters of acquired images 
3.2 Photogrammetric processing and visualization 
Within the previously finished pilot project the functionality of 
the automated DTM generation of several commercial digital 
photogrammetric stations (SocetSet, VirtuoZo, Image Station, 
Match-T) was tested. A comparison of the results with the 
manually measured reference showed that the automatic 
procedures do not bring required performance in this case. The 
height differences between automatically derived models and a 
manually measured reference rise up to 5-7 mm in a data set 
with a maximal elevation range of about 300 cm. However, 
using manual measurements, the theoretical accuracy of 0.3 mm 
(0.15%o of the flight height above the object) can be expected. 
Main reasons for the problems of commercial image matchers 
are distinct height differences and the low contrast in the relief. 
As mentioned before, a very precise DTM is needed (height 
error of 1 mm in maximum). Thus, the analogue images were 
selected as a basis for the photogrammetric processing and we 
distorted to manual measurements in the main project phase. 
Phototriangulation. The whole block was triangulated 
manually on an analytical plotter. The achieved accuracy of 
7.62 microns corresponding to 0.17 mm in object space is good 
enough. 
DTM generation. For the DTM generation 48 stereo pairs 
along the four strips as well as several additional image pairs in 
the mountainous area were measured manually. We used profile