wo thresh-
s in black.
cations of
the labels.
Machine -
allel Com-
Je Sparse
:nglewood
cation Ex-
chives for
9, Com. 3,
uordnung,
tion und
ammetrie,
3 C, 366.
res image
ct space,
ensing, 58
orrelation,
insing, 54,
ice recon-
ages. Pro-
m: Spatial
Computer
Academic
nted Mod-
perator for
Is and im-
n für das
he Kom-
ision) — À
gital pho-
mmission
rammetric
AUTOMATIC TIE POINT EXTRACTION IN AERIAL TRIANGULATION
Eija Honkavaara and Anton Hggholen
Finnish Geodetic Institute
Department of Photogrammetry and Remote Sensing
Geodeetinrinne 2
FIN-02430 Masala
Finland
e-mail. Eija.Honkavaara@fgi.fi & Anton.Hogholen @fgi.fi
Commission III - Theory and Algorithms
KEY WORDS: tie point extraction, aerial triangulation, adjustment, accuracy, automation, block
ABSTRACT
A new conceptual division of the automatic tie point measurement process into tasks is presented. An important feature in this
approach is that attention is paid to the accuracy questions and the treatment of problems in tie point extraction. A system for
automatic tie point measurement, which is under development at the Finnish Geodetic Institute, is outlined.
An empirical investigation on the number, distribution and completeness of the tie point observations was carried out. The OEEPE
test block Forssa with 30 um pixel size was used as test data. The investigated factors affected especially the height accuracy. The
accuracy of the block improved with an increasing number of observations, but only up to a certain limit. The 5x5 distribution of tie
point areas gave only slightly better accuracy than the 3x3 distribution. The accuracy of the block deteriorated with decreasing
completeness of the observations. The RMS errors in the check points were in the best case, when using automatic tie point
observations: X: 1.8 cm, Y: 2.3 cm and Z: 3.7 cm. The results are promising. There is reason to believe that the accuracy can be
further improved.
1. INTRODUCTION
During the last years, automation of the tie point measurement
process has been a popular research topic among
photogrammetrists. Quite a number of approaches with high
automation level have been developed.
The common idea of the systems seems to be as follows. First,
the areas where tie point extraction will take place are defined.
Corresponding points are measured in these areas using image
matching. Because the quality of the matched points is
unknown and may be poor, a huge number of tie points is
measured to achieve good accuracy. Interactive measurement is
used in problematic cases, which arise because of failures in
image matching.
There are questions concerning this approach. First, the level of
automation is questionable, because the rate of failures is non-
deterministic. Second, there are also uncertainty concerning the
accuracy and the reliability of a block. Previous results, see
(Honkavaara and Hggholen, 1995), show that the accuracy of
the block does not improve infinitely with an increasing number
of observations.
To handle these questions, the following conceptual division of
the tie point measurement process into tasks is suggested:
Tie point area definition.
Corresponding point definition.
Block adjustment and point selection.
Quality control.
Treating the unsuccessful tie point areas.
. Process flow.
These tasks are not necessarily separable, but may be combined.
QAM S ae
The research activities on automatic tie peint measurement have
been concentrated mainly on tasks 1 and 2. Anyway, it is clear,
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
that in order to make reliable, accurate and fully automatic tie
point measurements all of the tasks should be dealt with.
The tasks in the tie point extraction process are described and
the system for tie point measurement being under development
at the Finnish Geodetic Institute (FGI) is shortly outlined in
Section 2 (more thorough description is given in (Honkavaara
and Hggholen 1995)). Empirical results about the effect of the
number, distribution and completeness of the tie points are
presented in Section 3.
2. AUTOMATIC TIE POINT MEASUREMENT
2.1 Tasks in tie point measurement
2.1.1 Tie point area definition
It is usually sufficient to extract tie points in a limited number
of distinct, homologously distributed locations, so called tie
point areas. For stability reasons it is important to select tie
point areas so that there exists a maximal number of
overlapping images. For instance, in a typical photogrammetric
block with 6096 forward overlap and 20-3096 side overlap, 6-
fold overlaps are quite frequent. The tie point area distribution
is discussed in Section 2.2.1.
The paradox on defining tie point areas is that the exterior
orientation of the images as well as the topography should be
known. There exist different methods for tie point area
definition with varying complexity, see overview in (Fórstner
1995). The most simple methods use only approximate values
of the exterior orientations and assume smooth terrain. The
more sophisticated methods define the overlap areas by using
progressively refined image fingerprints (defined by using
orientation parameters and terrain heights determined during the
337
= EE
EORNM
Ncc REA