et al. 1994)
registration
olygons ex-
ans of edge
; carried out
n of the po-
a. Examples
's. Also this
laxation for
ips and im-
represented
these graphs
elations are
; successful-
ophisticated
dealt with.
5) use lakes
ous orienta-
. This work
(1991). Due
e of control
s a marginal
rientated are
ram of the
ms of area,
atellite orbit
matching is
ater bodies.
of the para-
nation. Suc-
, SPOT XS
been instal-
nd and the
dule for ab-
photo maps
-Westfalen.
.000. Three-
are used as
prior studies
989). Assu-
itation para-
n and +/- 1
; are projec-
ined. In this
area straight edges are extracted from the images. Mat-
ching between image and model edges is carried out via
cluster analysis. The orientation parameters are compu-
ted in a robust least squares adjustment based directly on
the matched edges. The approach has a built-in self
diagnosis and was successfully tested with more than 50
images. A critical point is the availability of the control
information in the form needed for the algorithm. How-
ever, with the growing popularity of city models three-di-
mensional wire frame models of buildings should be
increasingly available in the future. Houses represented
as wire frame models are also used for matching image
and object descriptions by Bejanin et al. (1994). Their
main emphasise, however, is on change detection, and
therefore no further explanations will be given here.
In summary, automatic absolute orientation has been
shown to work well in special applications. So far, only
single images have been processed. In contrast to interior
and relative orientation, hierarchy does not play a central
role here, since the control information is in general not
detectable across various image scales. Area based mat-
ching seems to suffer from the sensitivity of the grey
values due to different illumination and other disturban-
ces. Feature based and especially relational matching
have a much larger potential to solve the given task.
Redundancy can be exploited to some extend, if GIS
objects are directly used. In this case, the rather costly
signalization of specific targets becomes obsolete. It must
be ensured, however, that the GIS data are up-to-date,
and that their geometric accuracy is sufficient.
5 CONCLUSIONS
In this paper the state of the art of automatic image
orientation in photogrammetry was presented. It was
shown that interior and relative orientation can be and
have been cast into autonomous modules. Relative orien-
tation is also the core of automatic aerial triangulation.
Modules for interior and relative orientation and auto-
matic aerial triangulation are commercially available to-
day (Braun et al. 1996; Lue 1996; Madani 1996; de Vene-
cia et al. 1996). Most of these modules are implemented
in digital photogrammetric workstations (Heipke
1995a,b; ZPF 1995). As an alternative interior orientation
can also be implemented on a scanner workstation and
performed immediately after scanning. Using these mo-
dules an autonomous processing chain can be set up
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
starting from scanning the analogue photographs and
leading to epipolar images which can be viewed and
further processed in stereo (see also Sarjakoski 1990;
Mayr, Poth 1995).
The situation is somewhat different for automatic abso-
lute orientation. Although encouraging research results
and developments towards automatic systems exist for
special cases, a general solution is not feasible today.
Thus, semi-automatic measurement possibilities are of-
fered in most DPWS: the operator manually measures
control information (mostly points) in one image, and
matching is used to transfer this point into the other
images.
Despite the substantial improvements in automatic
orientation of photogrammetric imagery over the last few
years presented in this paper, the future of this area is
uncertain. This has to do with new sensor developments:
The reestablishment of the interior orientation is not
required any more as soon as digital cameras are used for
image acquisition. In addition, GPS and INS make the
indirect determination of the exterior orientation obso-
lete by providing direct measurements for the desired
parameters. Due to two reasons the absolute orientation
as described in this paper is expected to be the first step
vanishing from the photogrammetric processing chain:
(1) as mentioned the automatic extraction of control
information from imagery is very difficult and automation
in practice - while highly desirable - is not likely to occur
in the foreseeable future, and (2) using the more mature
GPS technology alone without having to rely on INS the
datum of an image block, and thus the absolute orienta-
tion parameters of all images can be determined. How-
ever, the described techniques developed for automa-
tically extracting control information from images can be
transferred to GIS data extraction and database revision.
Hence, new challenges remain to be tackled.
6 REFERENCES
Ackermann F., 1994: Practical experience with GPS supported aerial
triangulation, Photogrammetric Record (14) 84, 861-874.
Ackermann F., 1995a: Automatic Aerotriangulation, Proceedings, 2nd
Course in Digital Photogrammetry, Landesvermessungsamt
Nordrhein-Westfalen und Institut für Photogrammetrie, Univer-
sität Bonn.
Ackermann F., 1995b: Sensor- and data integration - the new challen-
ge, in: Colomina I., Navarro J. (Eds.), Integrated Sensor Orienta-
tion, Wichmann, Heidelberg, 2-10.
309
