ELSEVIER ISPRSJournalofPhotogrammetry& RemoteSensing52 (1997)122-131
PHOTOGRAMMETRY
& REMOTESENSING
Automatic aerotriangulation concept, realization and results
L. Tang a,*, j. Braun b, R. Debitsch b
a TechnicalConsulting and Software Developmentfor Photogrammetry,Remote Sensing and Geoinformatics, Riesstr. 65, 6.0G/332,
D-80993 Munich, Germany
bCarl Zeiss, D-73446 Oberkochen, Germany
Accepted8 April1997
Abstract
An automatic aerotriangulation system is described. Its design and development have been made to possibly meet every
requirement from photogrammetric practice. The system consists of five components, i.e. block preparation, fully automatic
tie point determination, semi-automatic control point measurement, interface to diverse block adjustment programs and
block post-processing. A relational data base takes care of communications among individual components. The fully
automatic tie point determination plays a key role in the whole system. Its realization follows the principle of image
connection and thus exhausts the technical potential to reach the highest level of automation. Intensive operational tests
were carded out using diverse image blocks with various configurations from photogrammetric practice. Different terrain
types and ground covering as well as various camera focal lengths and image scales were considered. On average 200-500
tie points were determined per image. The achieved standard deviation of image coordinates amounts to 0.2-0.5 pixel. The
computation time for the tie point determination is about 5 min per image. It was proven that automatic aerotriangulation
provides higher reliability of results and much more economy to the photogrammetric practice than ever before. The
system is now in daily operation in practice.
Keywords: automatic aerotriangulation; digital photogrammetry; digital image matching; tie point; image connection
1. Introduction
With the rapid development of digital technology,
photogrammetry is entering a digital era. The
most significant feature of digital photogrammetry
is the high automation of individual processing procedures,
e.g. automatic interior and relative orientation,
automatic digital terrain or surface modelling
and automatic orthoimage generation. This brings
much more economy to the photogrammetric practice
of today and leads, as a matter of fact, to a
*Correspondingauthor.
revolutionary change in the photogrammetric pro-
duction.
Aerotriangulation (AT) is an essential task in photogrammetry.
The tie point selection, transfer and
image coordinate measurement in the course of conventional
AT require intensive interaction of a human
operator and belong to the most laboured and timeconsuming
work. Thus, an automatic AT or AAT is
highly desirable in practice. Since the early nineties,
research and development of algorithms for automating
AT have been conducted at several institutions
(e.g. Tsingas, 1992; Schenk and Toth, 1993; Ackermann
and Tsingas, 1994). Today, AAT is one of
0924-2716/97/$17.00  1997ElsevierScienceB.V. All rightsreserved.
PII S0924-27 16(97)00012-9
L. Tanget aL/ISPRS Journal of Photogrammetry & Remote Sensing 52 (1997) 122-131 123
the highlights in the photogrammetric research and
development (e.g. Ackermann, 1995; FiSrstner, 1995;
Fritsch, 1995; Mayr, 1995; Schenk, 1995, 1996) and
commercial products of this kind are becoming available
on the market (e.g. Braun et al., 1996; Krzystek
et al., 1996; Miller et al., 1996).
PHODIS AT is such a commercial AAT system
from the Carl Zeiss company (cf. Braun et al., 1996).
Its design and development have been made to possibly
meet every requirement from the photogrammetric
practice (e.g. Tang, 1996). This paper deals
with issues concerning the concept and realization
of PHODIS AT. System components are described.
Achieved results are presented and economic issues
discussed.
2. Concept
2.1. AT steps
AT aims to determine exterior orientation parameters
of images of a photogrammetric block. It is usually
realized by a block adjustment using tie points
determined in the neighbouring images of the block.
A conventional indoor AT procedure can consist of
the following steps (cf. Fig. 1):
I Blockpreparation
4'
I Tiepointdetermination
4,
I GCPacquisitioninimages
I (Bundle)blockadjustment
~ '--'~Results~
Fig. 1.Flowchartofaerotriangulation.
(1) Block preparation, where among others images
are ordered according to the flight plan, camera
data and ground control information (e.g. ground
control points or GCPs) are collected.
(2) ~e point determination, which includes (a)
point selection, where distinct image points are chosen
around standard positions, marked and assigned
with unique names or number codes, (b) point transfer,
where selected image points are transferred to
the neighbouring images by means of e.g. a point
transfer device, (c) measurement of image coordinates
of tie points, which can be performed in mono
or stereo mode on a comparator or an analytical
plotter.
(3) GCP acquisition in images, where GCPs are
identified in the images with the help of given
sketches and the measurement can be done in the
same manner as that of tie points.
(4) (Bundle) block adjustment, which is carried
out by a computer program, using image coordinates
of tie points, image and object coordinates of
GCPs and camera data as input, and determines exterior
orientation parameters of the images and object
coordinates of the tie points.
2.2. Automation
Obviously, the block preparation requires human
knowledge and can hardly be automated. However,
using e.g. Global Positioning System (GPS) techniques
for the block flight may make this kind of
work much easier. The block adjustment is a purely
computational job and performed already without
any human interaction. Since specific knowledge is
needed to identify GCPs in images and GCP types
and shapes can vary considerably, a full automation
of the step of GCP acquisition in images is hardly
realizable. However, the measurement of image coordinates
can be overtaken by digital image matching
techniques (e.g. Gtilch, 1995). In addition, the
number of GCPs in a block is usually limited. Therefore,
a semi-automatic acquisition is very helpful
in practice. The tie point determination remains the
most labour- and time-consuming work in the whole
procedure. Since the determination of tie points in
images is a kind of non-semantic processing, i.e.
no specific features need to be recognized, a full
automation of this step is possible.
124 L. Tanget al./ISPRS Journalof Photogrammetry&RemoteSensing52 (1997)122-131
In one word, automation in AT means a fully automatic
tie point determination and a semi-automatic
GCP acquisition in images.
2.3. Design issues
Basic ideas for system design of PHODIS AT are
(a) to exhaust the technical potential for reaching the
highest level of automation in AT, and (b) to keep the
system requirements to a very limit.
In conception of the system, special attention was
paid to the following aspects:
* High reliability and robustness. This is the most
important issue for any automatic procedure. Therefore,
specific strategies are needed for algorithmically
realizing the automation.
* Easy operation. Only readily available information
and data of a block can be required. Human interaction
should be menu-driven. Automation means
really a 'one-clicking' action. No specific parameters
must be adjusted by users for automatic procedures.
* Any block configuration. The system should be
able to deal with any kind of block configuration
which may occur in practice.
* Any block size. Due to the huge amount of data
of a possibly large block and the limited storage
capacity of a computer, the processing can be carried
out only piece by piece. The system should be
capable of subdividing a block into small subblocks,
handling them one after another and adjusting either
the subblocks individually or the block as a whole.
* Open interface to block adjustment programs.
Since many kinds of commercial block adjustment
program packages or user-owned programs are already
in daily use of photogrammetric production, no
special block adjustment program must be included
in the system. Data exchange should be supported
for popular commercial program packages and also
open for any kinds of user-owned programs.
* Data format for plotting systems. Exterior orientation
parameters and/or tie points can be transformed
into diverse data formats for digital, analytical
and even analogue plotting systems.
3. System description
Fig. 2 shows the system structure of PHODIS AT.
Five procedures are involved in the system, i.e. block
preparation, fully automatic tie point determination,
semi-automatic GCP acquisition, (bundle) block adjustment
and block post-processing. A relational data
base holds all kinds of input information and data
as well as results of an AT block, and supports the
communication and the processing in the individual
procedures. The core of the whole system is the
procedure of fully automatic tie point determination.
Comprehensive descriptions on PHODIS AT can
be found in Braun et al. (1996). In the following,
Blockpreparation
4' 4,
Fully
automatic
tie point
determination
Semi-
automatic
GCP
acquisition
(Bundle)blockadjustment
~__._DATA BAS~.~
* Block information
* Imageinformation
* Cameradata
* Interiororientation
* GCPs
* Tie points
l~_~~Exterior orientation
Fig.2. SystemstructureofPHODISAT.
--~ Blockpost-processing
I digital,analytical, I
I analogue I
k-- _ _ plottingsystems /
L. Tang et al./ISPRS Journal of Photogrammetry & Remote Sensing 52 (1997) 122-131 125
key features and new updates of individual system
components are given.
3.1. Block preparation
The system requires the following readily available
inputs: (1) digital image data; (2) camera protocol;
(3) approximate projection centres; (4) ground
control information.
Approximate projection centres can be derived
either from the log file of a flight planning system
(e.g. T-FLIGHT) or simply by reading the flight plan.
GPS measurements, if available, can make this kind
of estimation more precise, but are not necessarily
required by the system.
Based on these input data and using a special
graphical editor, the block structure can be defined
and generated. A block can consist of one or more
normal and/or crossing strips and a strip of any number
of images. For an AAT system, the image data
amount is tremendous. A black-and-white aerial image
scanned with a geometric resolution of 15 /zm
and a radiometric one of 8 bits results in a data
amount of about 240 Megabytes. Therefore, it is
quite possible that in practice not all images of a
block can directly be accessed by computer. In this
case, the block must be subdivided into parts. A subblock
editor serves to define subblocks in a block for
independent processing. In order to guarantee a welltied
block for adjustment, subblocks should overlap
each other by at least two images in flight direction
and one image strip across the flight direction.
Before starting the automatic procedures, two basic
operations must be carried out for each image
in the subblock to be processed, i.e. image pyramid
generation and establishment of the image-pixel
coordinate relationship. The former is carried out
autonomously and the latter can be performed by an
integrated automatic or manual interior orientation
procedure.
A graphical block display helps to schematically
visualize the block, define or select a subblock and
optionally show the number, the icon or the interior
orientation result of each image in the selected
subblock.
3.2. Fully automatic tie point determination
This procedure is supported by a coarse-to-fine
image matching approach combining feature-based
matching (FBM) and least squares matching (LSM)
techniques. The approach is an extension of the one
which is successfully used for automatic relative
orientation in the digital plotter PHODIS ST from
Carl Zeiss (Tang and Heipke, 1993, 1994, 1996;
Tang et al., 1996).
3.2.1. Principle of image connection
The realization of fully automatic tie point determination
follows the principle of image connection
(cf. Tang, 1996). Instead of concentrating on standard
positions as usual by conventional methods,
the whole area of an image is searched for possibly
well-defined tie points. Two practical reasons support
this principle. First, tie points evenly distributed
over the whole area of an image intuitively present a
stable geometric connection to its neighbouring images
in the block. Second, the texture appearance on
an image is hardly foreseeable and concentrating on
certain small areas for good tie points is, therefore,
not very realistic for a fully automatic process.
As a matter of fact, the following evident advantages
come with this principle:
* Optimal use of information. Whatever an image
looks like, tie points will be extracted from those
areas where good texture exists. No prepositiouing
is needed and no special analysis of image content
is necessary. This leads to an optimal use of the
information provided by the image and the best
connection to the neighbouring images.
* High reliability and robustness. The optimal use
of information in an image also explores the best
possibilities for connection. Locally poor texture
will not affect the global connection. In addition,
an efficient mechanism based on some global mathematical
models (e.g. collinearity equations) can be
incorporated to ensure the geometric consistency of
the obtained tie points and, as a matter of fact, the
reliability and robustness of the final results.
* Global control. Global area image connection
instead of local point transfer can lead to a very
strong interior geometric stability of the block. The
block distortion can also be compensated optimally.
Thus, it can be expected that the number of GCPs
126 L. Tanget al./ ISPRS Journal of Photogrammetry & Remote Sensing 52 (1997) 122-131
BlockFormation
extract features
connect right image
connect previous strip
updatetie point list
+
Point Tracking
definewindows
performLSM
fillouttie point list
startlevel
I I t l
// ',i
/ ....
_ .... ,/,,.
intermediate level
!
s I
, \
I
I I
I I
I I
I I
I t l
..... |
' ; --,*
ie pom,u~, ) t / destinationlevel t
~ , ~ ............ _~t
Fig.3. Conceptoffullyautomatictie pointdetermination.
for the block adjustment might be reduced to a very
limit and even more labour-intensive work can be
saved.
* Easy operation. Precise initial exterior orientation
parameters of images are not necessary, and
neither are elevation models. Only the general block
information (e.g. strips, neighbourhoods or approximate
projection centres) is required to run the automatic
process.
3.2.2. Approach
In order to realize the principle of image connection
and at the same time to speed up the procedure
as much as possible from the practical point of view,
a hierarchical image matching approach was developed
based on image pyramid techniques. Fig. 3
shows the principle of fully automatic tie point determination
in PHODIS AT.
The image pyramids are divided into two parts
by defining a so-called intermediate level. The upper
part includes levels with small image sizes and lower
resolution and the lower part the ones with increasing
image sizes and higher resolution. Two steps are
involved in the fully automatic procedure. The first
step is called 'block formation', serving to connect
images of the whole block together through the
upper part of the pyramids till the intermediate level.
The second step is called 'point tracking', aiming at
achieving as high a measuring accuracy of each tie
point as possible through the lower pyramid levels.
The idea of introducing the intermediate level is
to arrive at an optimal combination of the use of
available information and the computational time.
Thus, the intermediate level will be defined at that
level, in which the tie point determination can still be
carried out fast enough and from which enough tie
points can be generated for a reliable point tracking.
Block formation is achieved by connecting individual
images with their neighbours in the block.
FBM is used to determine conjugate points in image
pairs. For each image, point features are first
extracted by means of an interest operator. They
are then matched according to certain geometric and
radiometric criteria. For connecting the right image
within a strip to the left one, the matched point pairs
are used as observations in a robust bundle adjustment,
in which relative orientation parameters and
object coordinates of the points are determined and
outliers are eliminated. The orientation parameters
and the point object coordinates are then forwarded
to the next lower pyramid level on which the matching
procedure is repeated. For connecting an image
of the previously processed neighbouring strip, due
to the possibly very limited overlap, a robust affine
transformation is performed to eliminate outliers in
the matching. Tie points are administrated in a list,
which is updated with new points during the course
of block formation. Manifold tie points are obtained
by checking those shared features of feature pairs in
common images. The final result of the block formation
is the tie point list generated at the intermediate
level, in which a tie point is provided with a unique
name, the number of tie images and a list of names
of these tie images including the measurements of
the image coordinates.
In order to precisely measure the image coordinates
of a point in the tie images, LSM is performed
pair by pair in all combinations through the lower
part of the pyramids. Around a given point pair at the
intermediate level, a reference and a search window
are defined. Six affine parameters and two radiometric
ones are calculated between the two windows.
For a convergent result, the cross correlation coefficient
between the two windows is then computed. If
the coefficient is larger than a threshold, the match
is declared as successful. An interest operator is
used in the reference window to find a proper point.
L. Tanget al./ISPRS JournalofPhotogrammetry&RemoteSensing52 (1997)122-131 127
This point is then transformed to the search window
via the afline parameters, defining the corresponding
point there. These two points are mapped onto the
next lower pyramid level and LSM repeats. At the
end of point tracking, the tie point list is updated
with image coordinate measurements in the original
image resolution. Since the number of tie points in
an image can be unnecessarily large for the block
adjustment, the 2-fold tie points are tracked only
selectively.
As a result, the fully automatic procedure for tie
point determination in PHODIS AT delivers those tie
points which are evenly distributed over the whole
area of each image and accurately measured in original
images.
With respect to the issue of implementation, no
control parameters for matching need to be set by
the user. Thus, this procedure can be considered an
autonomous process.
3.3. Semi-automatic GCP acquisition
In PHODIS AT, the measurement of GCP image
coordinates can be carried out manually or semiautomatically.
The semi-automatic GCP acquisition
in images leaves the identification task to the human
operator and the fine coordinate measurement
to the LSM algorithm. The idea is that tie points
evenly distributed over the whole area of an image
are used to define local transformation parameters
to the neighbouring images. Given a position in the
image, the approximate corresponding positions in
the neighbouring images can be calculated via the
local transformation parameters. In this way, the human
operator needs to identify and finely locate a
GCP only in one reference image, and a hierarchical
multi-image LSM algorithm then takes care of
finding those correspondences in other images. Supposing
that a GCP appears in six images, all six
images have to be searched by the human operator
for the GCE Thus, even more comfort is provided
by PHODIS AT for this task after the automatic
procedure of tie point determination.
3.4. (Bundle) block adjustment
The block adjustment in PHODIS AT is open
for any popular commercial program packages or
user-owned programs. This intention is based on the
fact that the photogrammetric practice is accustomed
to its own block adjustment programs since decades
and it is not necessary to convince anyone to use an
unfamiliar program. Interfaces need to be created for
data exchange between PHODIS AT and user-owned
systems, which can even be started on any kind of
computer platforms.
3.5. Block post-processing
The block post-processing includes documentation
of GCPs and new points (GCPs-to-be), and
generation of model setup parameters for image orientation
on digital, analytical and even analogue
plotting systems.
4. Practical experiences and economic issues
During the whole phase of development, PHODIS
AT was intensively tested. New requirements and
suggestions from various testsides considerably
helped to improve and complete the functionality
and performance of the system. In the following, we
will briefly present our experiences with PHODIS
AT and discuss some economic issues with respect
to AAT.
Table 1 shows the results obtained by using
PHODIS AT. All the test blocks came from photogrammetric
practice and present various terrain
types (e.g. flat, hilly) and ground covering (e.g.
snow/ice, urban or industrial areas, forestry). Images
were taken by cameras with focal length ranging
from 85 mm to 305 mm at scale factors of I : 3000 to
1 :72,000, and scanned in a resolution of 15, 20 and
28 /zm. On average 200-500 evenly distributed tie
Table 1
Practicalresultsobtainedby usingPHODISAT
Resolution Focallength Scalefactor o0
(/zm) (ram) (/zm)
Antarctic 15 85 1:72000 7.7
Urban 15 305 1:4000 5.2
Hilly/forested 15 153 1:3000 5.1
Rural 20 153 1:5500 3.9
Industrial 28 153 1:3000 5.9
Poorcontrast 28 153 1:3000 6.0
128 L. Tang et al. I ISPRS Journal of Photogrammetry & Remote Sensing 52 (1997) 122-131
~i~ ~
I
7\
7
/
Fig. 4. Three examples of 3-fold tie points automatically determined by PHODIS AT in images of urban areas.
points were determined per image, which guarantees
a high reliability of the final result. The achieved
standard deviation of image coordinates amounts to
0.2-0.5 pixel, a level of accuracy that a human operator
can reach. The computation time for the tie
point determination is about 5 min per image on a
Silicon Graphics workstation with R4400 processor
(200 MHz).
Fig. 4 gives some typical examples of tie points
automatically determined by PHODIS AT in images
of urban areas. Fig. 5 shows the final distribution of
tie points on a selected part of a strip of infrared images
taken in a coastal area to demonstrate how well
the principle of image connection for automatic tie
point determination works in the case that no suitable
texture can be found in the standard positions.
Two examples of practical photogrammetric
blocks with crossing flights are given in Fig. 6.
Obviously, the connection of crossing strips with
normal ones should not be restricted to any predefined
positions on images. Again, the principle of
image connection brings another evident advantage
to make this kind of strip connection straightforward.
Fig. 7 gives an example of a 6-fold tie point found
L. Tanget al./ISPRS Journal of Photogrammetry & Remote Sensing 52 (1997) 122-131 129
Fig. 5. The PHODISATtie pointdistributionon a selectedpartof a stripof infraredimagestakenin a coastalarea,wheretie pointsare
markedby 'whiteplus' and darkhomogeneousareaspresentwatersurfaces.
between a normal and a crossing strip of block (b) in
Fig. 6.
Table 2 presents an empirical time comparison
between the analytical AT and PHODIS AT for
processing a block consisting of 110 images scanned
in 14/zm and 30 GCPs. The analytical AT needs 140
hours or about 18 days (8 h/day) of human work.
On the contrary, only 2 days are needed when using
PHODIS AT by the human operator and another
two days by the computer, which do not necessarily
belong to the standard cost accounting in general.
This means that PHODIS AT is more economic than
the analytical AT by a factor of 9.
Using AAT brings not only a considerable time
improvement but also more comfort for the human
worker. The role a human operator plays is mainly
to supervise or control individual procedures. The
hardest work is then taken over by the computer.
5. Conclusions and outlook
Commercial systems for AAT become available
on the photogrammetric market. Their practical use
brings much more economy and leads to a revolutionary
change in the photogrammetric production.
In the photogrammetric practice of today, which is
still dominated by analytical or even analogue systems,
AAT is ready to replace the conventional AT
to directly provide image orientation to the analytical
or analogue systems. With rapid advances of
digital photogrammetry, the vision that a production
chain can be built up which allows automatic image
scanning, automatic aerotriangulation, automatic
digital terrain or surface modelling and orthoimage
generation, and even automated data acquisition for
geographic information systems, is becoming reality
(Tang, 1997).
130 L Tanget al./ISPRS Journal of Photogrammetry & Remote Sensing 52 (1997) 122-131
Fig. 6. Two examples of practical photogrammetric blocks: (a) block with perpendicular crossing strips; (b) block with arbitrarily angular
crossing strips.
Table 2
Comparison between the analytical AT and PHODIS AT for a block of 110 images scanned in 14/xm, including 30 GCPs
Analytical AT PHODIS AT
procedure human (h) procedure human (h) computer (h)
Scanning 2 18.4
Preparation, Preparation 2 11
tie point determination, 130 tie point determination 9.2
GCP acquisition GCP acquisition 3
Post-processing 10 Post-processing 8
Data backup 1 7
Total 140 Total 16 45.6
References
Ackermann, E, 1995. Automatic aerotriangulation. Proceedings
of 2nd Course in Digital Photogrammetry, Bonn.
Ackermann, F., Tsingas, V., 1994. Automatic digital aerial triangulation.
Proceedings ASPRS/ACSM Annual Convention,
Reno, Vol. I, pp. 1-12.
Braun, J., Tang, L., Debitsch, R., 1996. PHODIS AT -- an automated
system for aerotriangulation. Int. Arch. Photogrannn.
Remote Sensing 31 (B2), 32-37.
L. Tang et al. / ISPRS Journal of Photogrammetry & Remote Sensing 52 (1997) 122-131 131
Fig. 7. An example of a 6-fold tie point found between a normal {4759, 4760, 4762} and a crossing strip {4737, 4736, 4735} of block
(b) in Fig. 6.
Frrstner, W., 1995. Matching strategies for point transfer. In:
Fritsch, D., Hobbie, D. (Eds.), Photogrammetric Week '95.
Wichmann, Karlsruhe, pp. 172-183.
Fritsch, D., 1995. Introduction into digital aerotriangulation. In:
Fritsch, D., Hobbie, D. (Eds.), Photogrammetric Week '95.
Wichmann, Karlsruhe, pp. 165-171.
G01ch, E., 1995. Automatic control point measurement. In:
Fritsch, D., Hobbie, D. (Eds.), Photogranunetric Week '95.
Wichmann, Karlsruhe, pp. 185-196.
Krzystek, P., Heuchel, T., Hirt, U., Petran, E, 1996. An integral
approach to automatic aerial triangulation and automatic DEM
generation. Int. Arch. Photogramm. Remote Sensing 31 (B3),
405-414.
Mayr, W., 1995. Aspects of Automatic Aerotriangulation. In:
Fritsch, D., Hobbie, D. (Eds.), Photogrammetric Week '95.
W1chmann, Karlsruhe, pp. 225-234.
Miller, S.B., Paderes, EC., Walker, A.S., 1996. Automation in
digital photogrammetric systems. Int. Arch. Photogramm. Remote
Sensing 31 (B2), 250-255.
Schenk, T, 1995. Zur automatischen Aerotriangulation. Z. Photogramm.
Fernerkundung (ZPF) 63, 137-144.
Schenk, T., 1996, Digital aerial triangulation. Int. Arch. Photogramm.
Remote Sensing 31 (B3), 735-745.
Schenk, T., Toth, C., 1993. Towards an automated aerial triangulation
system. ASPRS/ACSM Annual Convention, Reno, pp.
652-660.
Tang, L., 1996. Toward automatic aerotriangulation. Proceedings
of Geoinformatics '96, Wuhan, Vol.One, pp. 433--440.
Tang, L., 1997. Towards an automated photogrammetric system.
Proceedings of GIS AM/FM ASIA'97 and Geoinformatics '97,
Taipei, Taiwan, May 26-29, 1997.
Tang, L., Heipke, C., 1993. An approach for automatic relative
orientation. In: Griin, A., Kahmen, H. (Eds.), Optical 3-D Measurement
Techniques, II. Wichmann, Karlsruhe, pp. 347-354.
Tang, L., Heipke, C., 1994. An Automatic procedure for the relative
orientation of aerial images. Proceedings of the 3rd International
Colloquium of LIESMARS on Integration, Automation
and Intelligence in Photogrammetry, Remote Sensing and GIS,
pp. 49-58.
Tang, L., Heipke, C., 1996. Automatic relative orientation of
aerial images. Photogramm. Eng. Remote Sensing 62 (1),
47-55.
Tang, L., Poth, Z., Ohlhof, T., Heipke, C., Batscheider, J., 1996.
Automatic relative orientation -- realization and operational
tests. Int. Arch. Photogramm. Remote Sensing 31 (B3), 843-
848.
Tsingas, V., 1992. Automatisierung der Punkt0bertragung in
der Aerotriangulation durch mehrfache digitale Bildzuordhung.
Deutsche Geod~itische Kommission, Reihe C, Nr. 392,
Miinchen.