)N
transfer and
may be the
ns, such as
en accepted
V) extraction
Jlation using
nd computer
selection and
angulation in
5 um and 30
n Intergraph
; the INPHO
ised on this
ogram using
edures. lt is
time of less
has opened
volution has
10logies and
principles of
s have. One
y the digital
r automated
nply did not
ient. Today,
workstations
ese digital
accepted as
le mapping
nodel (DTM)
ie automatic
jues are the
new digital
ker, Madani,
otry involves
je matching
and transfer
programs for
loped.
for digital
ategy, which
be classified
as operator-guided automatic point transfer. The
operator selects a suitable tie point on one or more
overlapping images using a measuring mark. At this
time, image coordinates of the tie point are recorded.
Next, this tie point is transferred to other overlapping
images either interactively using mono or stereo
observations or automatically using image matching
techniques. The operator has full control of the operation
and produces very accurate results with this approach;
however, the approach is time consuming and requires a
skilled operator (Ackermann, Tsingas, 1995). A
performance time of about 8 minutes per image was
achieved on a digital triangulation of 979 aerial
photographs using an Intergraph digital photogrammetry
system (Beckschafer, 1995).
The second strategy is a fully automated aerial
triangulation. The main objective of this strategy is to
automate point transfer and tie point measurement
operations. Therefore, it minimizes manual work and
reduces operator intervention as much as possible.
Up to now, three different methods for automatic point
transfer and measurements have been reported. These
methods use either feature-based or least squares based
matching techniques (Tsingas, 1994; Schenk, 1993;
Ackermann, 1995; Krzystek, 1995).
Tsingas (1994) uses feature-based matching in three
levels of image pyramids. Feature points are extracted
in every digital image, and preliminary pairwise matching
is performed by computing correlation coefficients.
Then, gross errors are detected and eliminated by using
an affine transformation model. Finally, a graph-
theoretical model is used to overcome the problems of
small matching errors and ambiguities.
Tsingas’ approach was tested on OEEPE block
FORSSA. Twenty-eight aerial (4 strips of 7 photos)
photographs of scale 1:4000 were scanned by
PhotoScan at 15 um and 30 um resolutions, and three
runs were made. Results of these runs were reported by
Ackermann and Tsingas (1994). As indicated in this
report, a O, value of 6.2 um corresponding to 0.41 pixel
size was obtained. A performance time of achieving
bundle adjustment at the highest level of the image
pyramid was about 5.3 minutes per photo on a Silicon
Graphics IRIS Indigo Workstation (3000 MIPS, 33 MHZ).
In a recent test, the same project was triangulated on a
Silicon Graphics Indigo 2 workstation (4400 MIPS, 200
MHZ), and a performance time of about 1 minute per
image was achieved (Fritsch , 1995).
Schenk (1993) uses a very comprehensive, integrated
matching concept including point and edge feature
matching. This system consists of three modules. The
first module generates a photo mosaic, including DTM
points and footprints of all images at a coarse resolution.
The second module selects block points at strategic
locations using information obtained in the first step. The
third module uses a least squares multi-image matching
in object space to determine conjugate points as
accurately as possible (Toth, 1993).
Krzystek (1995) uses a feature-based | matching
approach and aims at an operational system leading to
very reliable and accurate bundle adjustment results.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
The intention of the concept is a fully automatic
procedure that includes the initialization and the
automatic block adjustment. Only at the beginning an
initial parameter setup, the interior orientation, and the
interactive measurement of control points are needed. If
necessary, the results of the automatic block adjustment
can be analyzed and post edited. The approach
comprises point selection, point measurement, point
transfer, and block adjustment ion a single process.
The success of any image matching technique is
dependent on the availability of “good” approximate
values for exterior orientation parameters. Whereas
some of these parameters are known approximately
(photo base, average terrain elevation, and average
flying height), others are not available with sufficient
accuracy. The uncertainty in image matching is due to
maintaining a proper sidelap and the amount of terrain
relief with respect to flying height. The side lap problem
may be solved by careful flight mission planning and by
using proper navigational tools (Fórstner, 1995). Modern
navigational tools, such as Global Positioning System
(GPS) and Inertial Navigation System (INS), will be
standard in future aerial photography. These tools not
only control the flight mission but also provide good
approximate values for all exterior orientation
parameters. In addition, they will reduce the number of
costly ground control points needed in the conventional
aerial triangulation. For more information about
characteristics of these three automatic digital
triangulation strategies, refer to Förstner (1995).
The uncertainty in obtaining good approximate values
due to the terrain relief may be solved by increasing the
amount of overlaps and by using or creating an
approximate DTM. The automatic point transfer
techniques use image pyramids and coarse to fine
resolution image matching operations. The matched
image points, exterior orientation parameters, and object
point coordinates resulting in a low level are used as
approximate values for the next level in the image
pyramids.
In the following sections, brief descriptions about the
digital photogrammetry hardware and software used are
given. Operational procedures of a semi-automatic and
an automatic aerial triangulation are compared using two
different photogrammetric data sets. On the basis of this
study, some preliminary results and accuracy indicators
are reported. Finally, general concluding remarks on the
accuracy and performance of an automatic aerial
triangulation are provided.
2. MATERIALS, INSTRUMENTS, METHODOLOGIES
Two small blocks of aerial photographs with different
photo scales, cameras, and types and numbers of
control points were selected from larger projects. The
Texas project is a very accurate GPS photogrammetry
test field established by the Texas Department of
Transportation. — Signalized control points were spaced
about 90 m apart. Flight design and control point
spacing were such that each model had between 15 to
20 control points. Twenty-one diapositives (strips of 7
photos) from the Texas project were selected for this
study. The Maryland project is a high-precision GIS
