progress
digital
GPS
=
analytical
i i i i i >
1 T T T T » time
1960 1970 1980 1990 2000
Figure 1: Past and predicted development of aerial
triangulation.
use of modern aerial triangulation techniques in pho-
togrammetric practice could be represented by an-
other curve. It would be much flatter and shifted to
the right, accounting for the typical time shift that
can be observed between inventions, availability of
operational systems and their widespread use.
The rapid progress in the late sixties and early seven-
ties was mainly in the areas of block adjustment meth-
ods and error propagation analysis. With the avail-
ability of analytical plotters some procedural changes
took place. For example, one of the most significant
results was the great simplification of the point trans-
fer problem. GPS had, and is going to have, a great
impact on aerial triangulation. Not only did it spark
new research interest, but it also changed significantly
the way aerial triangulation projects are carried out.
Once a sensitive problem, the determination of good
control points is now a snap. Using differential GPS
methods during data acquisition greatly reduces the
number of control points. Digital photogrammetry
and computer vision will bring about new changes
that will shape aerial triangulation as dramatically
as analytical photogrammetry did during the past 30
years.
Not only is aerial triangulation in an exciting state of
change, but the entire field of photogrammetry. It is
safe to predict that the majority of papers presented
at the XVIII ISPRS Congress will deal in one way or
another with digital photogrammetry. As the name
suggests, digital photogrammetry deals with digital
imagery and the goal is to capture, to store, and to
process images automatically. Ever since computers
became available, researchers tried to mimic the men-
tal faculty of seeing. It is tempting to endow comput-
736
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
ers with the information processing capabilities sim-
ilar to those of human operators. How soon can we
expect a machine that produces maps automatically,
performs an aerial triangulation, generates DEMs and
orthophotos at the touch of a button? The answer
almost entirely depends on how well the human pro-
cesses necessary to solve photogrammetric tasks are
understood. There is common agreement in the artifi-
cial intelligence, computer vision, and photogramme-
try communities that these processes are poorly un-
derstood. It appears that the broader the knowledge
that is required to solve a problem, the more difficult
it is to encapsulate it. For example, to make a map
relies on a much broader knowledge base than aerial
triangulation. The major difference is in the type of
knowledge: the map making process requires common
sense, general “world knowledge” and image under-
standing abilities; aerial triangulation, on the other
hand, uses a great deal of very specialized knowledge
(block adjustment, error propagation, etc.). This is
good news because the knowledge of an aerial triangu-
lation expert is far less difficult to represent explicitly.
This warrants the assumption that aerial triangula-
tion can be automated much more readily than, say,
making maps. Oddly enough, it is much more diffi-
cult to explain aerial triangulation to a layman than
how to make a map.
This paper does not review existing digital aerial tri-
angulation systems nor is it a status report. Rather
it elaborates on the fundamental problems-—called es-
sential tasks here—-that are inevitably linked to au-
tomatic aerial triangulation. The next section com-
ments on terminology and attempts to classify digi-
tal aerial triangulation and to contrast it to existing
procedures. The paper focuses on automatic aerial
triangulation. Readers familiar with the subject may
skip the introductory sections and proceed directly to
Section 5, where essential tasks are derived and pos-
sible solutions are analyzed. Finally, the paper con-
cludes with some remarks on the consequences digi-
tal aerial triangulation will have on photogrammetry.
Incidentally, if some opinions and ideas expressed in
this paper appear provocative to the reader, then it is
purposely in the hope of stimulating interesting dis-
cussion.
2 BACKGROUND
Digital aerial triangulation enjoys great interest in
digital photogrammetry, both by researchers and
users. This is manifest in many publications. The
organizers of the Photogrammetric Week '95 devoted
one session to the subject, featuring 10 interesting
contributions that addressed various aspects of digi-
tal aerial triangulation (see References).
2.1 Taxonomy
Whenever new technologies emerge usually no univer-
sally accepted terminology exists in the beginning and
int
the
foll
def
me!
The
ten
2.3
In a
follo
Pre
Poir
Poir
