constrained
om stereo—a -
nd Cybernet-
mage match-
using aerial
, Jr., editors,
Scene Ana-
do, FL, April
ystem. Inter-
note Sensing,
Object Space
t of Geodetic
rsity, Colum-
alues as un-
g. Submitted
atching tech-
etry and Re-
rated aerotri-
SPRS Annual
ns, LA, 1993.
ation and res-
Photogram-
. In Press.
e reconstruc-
togrammetry
An Integral Approach to Automatic Aerial Triangulation and Automatic DEM
Generation
Krzystek P., Heuchel T., Hirt U., Petran, F.
INPHO GmbH, Stuttgart
email: inpho@inpho.de
Commision III, Working Group 1II/2
KEY WORDS: Automatic aerial triangulation, automatic DEM generation, multiple image matching
ABSTRACT:
Automatic aerial triangulation will push the economical efficiency of digital photogrammetry to a new level since the
time consuming block preparation is avoided and costly interactive input is substantially reduced. Automatic aerial
triangulation appears as a turnkey method if it is combined with an integrated DEM generation process, which takes
full advantage of the multiple image overlap instead of the standard stereo overlap.
The paper presents a new approach to automatic aerial triangulation that comprises the automatic derivation of tie
point areas (7 Gruber point positions) and a hierarchical matching of tie point clusters as well. The kernel of the
system is an integrated bundle solution which robustly matches feature points in multiple images and estimates
orientation parameters simultaneously. The approach comprises point selection, point measurement, point transfer
and block adjustment in one process. DEMs are generate
d either locally in the tie point areas or for the entire block
rigorously using the multiple image overlap wherever possible.
The paper outlines the approach for automatic aerial triangulation and reports on practical results assessed in several
controlled tests. The examples given include the initialization part of the system and the automatic block adjustment
as well. The results for 30 um pixel size indicate that the approach leads to excellent results in the block adjustment,
which meet at least the accuracy requirements of a standard aerial triangulation. The computation time of the entire
automatic block adjustment is better than 6 minutes per image. The benefit of the integrated DEM generation is
demonstrated with the OEEPE FORSSA test block by comparing the results of the block DEM against the theoretical
results of the automatic block adjustment. Preliminary results indicate a DEM accuracy of 0.1 %o of the flying height.
1. Introduction
Digital photogrammetry is on the way to reach a new
level of economical efficiency because of an ongoing
improvement of digital systems towards a higher level of
automation, accuracy and product quality. System
providers make great efforts to embed fully automatic
procedures on their digital systems in combination with
powerful editing and visualization techniques. Automatic
DEM generation, orthophoto production, and the
automation of the interior and the relative orientation
have meanwhile become a kind of a standard of today's
digital systems. Automatic aerial triangulation in
particular is subject to rapid practical development, since
it is one of the keys to the economical breakthrough of
digital photogrammetry. It is undoubted, once the aerial
triangulation has successfully reached an automatic level
in a digital production environment, that project costs will
be significantly reduced.
Several approaches have been suggested from the
scientific community and from some system providers as
well. We have to mention the important developments of
Toni Schenk's group at the Ohio State University
(Schenk, Toth, 1993; Toth, Krupnik, 1994;Schenk, 1995)
and the solutions of the Stuttgart University (Ackermann,
Tsingas, 1994; Fritsch, 1995) Mayr (1995) and
DeVenecia et. al. (1996) are reporting about commercial
systems with high automation level. The related
approaches differ quite significantly in detail, mainly with
respect to the matching strategy. However, they all have
in common, that they are intended to operate fully
automatically.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
The subject of this paper is an approach to automatic
aerial triangulation which is currently carried out at
INPHO company as a software development (MATCH-
AT). It was presented in its key ideas in Krzystek et. al.
(1995). Heuchel et. al. (1996) reports very recently about
first practical results. The approach provides tie point
clusters by means of the feature-based matching
technique. The images of a block are orientated
simultaneously with the help of an integrated bundle
adjustment that robustly gets rid of mismatches. The
geometric matching criterion becomes the ray
intersection at homologous points instead of the widely
used affine transformation. The approach comprises the
point selection, point measurement, point transfer and
the block adjustment in one single process.
In addition, a DEM is created either locally at the tie point
areas or for the entire block area. The concept of an
integrated DEM generation is particularly important. As
far as the automation of the aerial triangulation is
concerned, it helps to principally overcome the problem
of relief displacement in the initialization part in case of
large scales and hilly or mountainous terrain.
Furthermore, the matching of image patches is more
effective, if the terrain surface is considered by an
appropriate resampling. Once a DEM generation has
been integrated in such a system, it is very easy to derive
a DEM for the entire block. Such a DEM has the benefit
of high quality and accuracy since the multiple image
overlap is rigorously used wherever possible. It should be
recalled that in a normal block with 60 96 end lap and 20
96 side lap only about 25 96 of the image area is covered
405
