tanbul 2004 
ACCURACY OF AUTOMATED AEROTRIANGULATION AND DTM GENERATION 
FOR LOW TEXTURED IMAGERY 
M. Sauerbier 
ETH Zurich, Institute of Geodesy and Photogrammetry 
Swiss Federal Institute of Technology Zurich 
ETH Hoenggerberg, CH-8093 Zurich, Switzerland 
www.photogrammetry.ethz.ch 
martin.sauerbier@geod.baug.ethz.ch 
IC WG IVIV 
KEY WORDS: Accuracy, Adjustment, Aerial, Automation, DEM/DTM, Matching, Processing 
ABSTRACT: 
The Nasca-Project at ETH Zurich aims for a GIS-based analysis of the topography and the geoglyphs carved into the ground by the 
ancient Nasca (200 B.C. — 600 A.C.) in the desert region of Nasca/Palpa, about 500 km south-east of Lima, Peru. In 1998, three 
blocks of aerial images have been acquired during a photoflight. Two of these blocks (about 400 images, scale 1:7.500) were 
processed using an analytical plotter Wild S9 during 4 years of manual measurements, resulting in a high resolution DTM and 3D- 
vectors of the geoglyphs as well as topographic elements (rivers, streets, houses). Especially aerotriangulation and DTM- 
measurements have been time-consuming processes. For the third block of Nasca imagery (about 400 images at a scale of 
approximately 1:9100) we aim for an automated processing (Aerotriangulation and DTM generation) to provide accurate data as a 
basis for terrain analysis (visibility studies, surface calculation etc.) in relation to the geoglyphs (Grün et al., 2003). 
Actual digital photogrammetric stations (DIPS) are examined for this case of low textured imagery due to the desertous characteristic 
of the landscape. Aerial triangulation and DTMs generated using Z/I's Image Station™ modules ISDM and [SAE, version 4.00, and 
Supresoft Inc. Virtuozo™ versions 3.1 and 3.3 are compared to reference data measured on an analytical plotter Wild S9. The main 
focus of this paper is on DTM generation, as for aerial triangulation no strictly comparable results, produced fully automatically, 
could have been achieved. 
60%. See figure | for an example of the used images and their 
content. 
The landscape covered by the images is mostly flat, here and 
I. INTRODUCTION 
One of the most important limiting factors concerning 
automated aerotriangulation and DTM calculation is texture. 
Actual digital photogrammetric stations allow for good results 
in most cases if the processed imagery contains texture with 
high contrast. Problems occur processing images which cover 
low textured areas, e.g. forests, glaciers, grasslands or deserts. 
Especially in these areas, matching algorithms fail measuring 
identical points in two or more images which leads to blunders 
or even non-measured points. The consequences during 
aerotriangulation can be instable relative orientations of images 
if there are not enough points to achieve a regular distribution, 
and therefore a decreased accuracy of the absolute orientation. 
DTM generation is affected by blunders or even gaps in areas 
where matching is impossible (Baltsavias et al., 1996). Both 
cases require costly manual editing or measurement. The data 
used for the examination of aerotriangulation consists of a 
subblock of 26 aerial images (B/W) of the Pampa de Nasca 
which were triangulated on an analytical plotter Wild S9, using 
IGP's bundle adjustment software BUN and on Virtuozo 3.1 
during a diploma thesis (Keller, 2003). These images were then 
also processed on Z/I's Image Station. To compare DTM 
generation, one stereo model of this subblock and 4 stereo 
models of another subblock were used. For both 
aerotriangulation and DTM generation the results of the 
automated processing were compared, no manual measurements 
or editing were applied. The area covered by the generated 
DTMs does not contain vegetation or buildings, thus the DSMs 
calculated by the DIPS can be taken as DTMs. 
The used images were scanned at a resolution of 15 microns 
according to 13.65 centimetres in object space. The overlapping 
along strip as well as perpendicular to the flight direction is 
there interrupted by quebradas, usually dry valleys formed by 
draining water from the Andes. The surface consists of little 
stones and sand.