atellites, a
flight of 4
at all the
the usual
wns to be
arameters.
‘the flight.
ambiguity
is linear
timated for
rs have to
is lost.
ly correct,
on linear
n iterative
BINGO-F,
additional
ments for
orithms.
1ethod will
easons as
etric block
period for
ysteme für
gen mit
schaftliche
wesen der
1 1983.
c Compu-
idle Block
ptical 3-D
ober 1995
3PS-Beob-
onsbestim-
chrichtung
nover Nr.
DO IT ON THE GROUND: INCREASING RELIABILITY AND ACCURACY OF AUTOMATIC
AEROTRIANGULATION BY MATCHING IN THE OBJECT SPACE
Amnon Krupnik
Department of Civil Engineering
Technion-Israel Institute of Technology
Haifa, 32000 Israel
krupnik@tx.technion.ac.il
Commision Ill, Working Group 2
KEY WORDS: Automatic aerotriangulation, Matching, Object Space, Multiple-Patch, Least-Squares
ABSTRACT
Automatic aerotriangulation is one of the major topics currently studied by the photogrammetric research community. Its
motivations are the potential increase in efficiency, accuracy and reliability of the aerotriangulation task. The work described
here is concerned with the last step of automatic aerotriangulation, which is automatically matching conjugate points from
more than two images. The matching strategy proposed in the paper is multiple-patch matching in the object space. lt is based
on iterative reconstruction of small object surfaces around tie points, and matching multiple image patches simultaneously. In
this paper, the idea is described, and experimental results are shown and discussed.
1 INTRODUCTION
Aerotriangulation has been a common photogrammetric tech-
nique for obtaining exterior (and possibly interior) orientation
parameters for a set of aerial photographs. Its main purpose
is to reduce the number of ground control points required
for orienting these photographs. Rather than a few ground
control points for each stereo model (minimum three), such
points are required less often, which reduces the cost of a
photogram metric mapping project considerably.
One of the major tasks in aerotriangulation is the measure-
ment of conjugate points on two or more partially overlapping
photographs. These points tie the photographs to each other.
With the current trend towards digital photogrammetry, and
the use of softcopy photogrammetric workstations, there is
a growing interest in automating the aerotriangulation task
[Ackermann (1995)], and particularly the measurement of tie
points. This interest is motivated by three key factors:
e Efficiency: Manual aerotriangulation, carried out by a
human operator, is a time consuming procedure. Con-
ducting it automatically in a batch mode, or even in
a semi-interactive environment will save many human
working hours.
e Accuracy: Hardcopy photographs are usually superior
to digital images in terms of resolution. Automatic
measurement using sub-pixel algorithms, compensates
for the inferiority of the digital images, and also allows
simultaneous measurement on more than two images,
which is not possible for a human operator.
e Reliability: Compared to a human operator, an auto-
mated system is capable of finding a large number of tie
points during relatively short time. This in turn leads
to more reliable results since the redundancy is large
and blunders are detected easily.
Research studies describing ways for automating the aero-
triangulation procedure were presented in [Helava (1988),
Ackermann and Tsingas (1994), Agouris (1992)] (see
[Krupnik (1994)] for a more detailed description). “The
most comprehensive and sophisticated concept for fully
automatic aerial triangulation..." [Ackermann (1995)] is the
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996
Automatic Aero Triangulation System (AATS), presented in
[Schenk and Toth (1993), Toth and Krupnik (1996). The
last phase of AATS, accurate, multiple-patch matching is the
main concern of this paper.
Although accurate matching has been the sub-
ject of numerous research studies in the fields of
photogrammetry and computer vision (see eg,
Baltsavias (1991), Dhond and Aggarwal (1989),
Lemmens (1988), Wrobel (1988) for reviews), its ap-
plication for aerotriangulation adds some new aspects that
have not been fully addressed:
e The exterior orientation parameters are not known.
Therefore, constraining the search space for the match-
ing by geometric conditions, e.g., the epipolar line, is
not possible.
e The accuracy requirements are more demanding than in
other applications (e.g., automatic generation of DEM).
e Only a single point is matched in each area. Therefore
the matching should be sufficiently reliable, and not de-
pendent upon the surrounding surface for determining
wrong solutions.
e More than two overlapping image patches are used at
each matching location. In order to obtain consist-
ency, a simultaneous matching of more than two image
patches is required.
These aspects are addressed and considered in the multiple-
patch matching in the object space, presented in this paper.
The method is based on hierarchically reconstructing the elev-
ations of small surface patches, centered around each matched
tie point. Having these surface patches, image patches (two
or more) are warped and simultaneously matched. Each itera-
tion of the algorithm improves both the calculated orientation
parameters and the surface patches around each point.
In the following sections, the motivations for the proposed
algorithm are explained, the algorithm is shown, and experi-
mental results are presented and discussed.
399
ee me re EE
mere
rn
Er SES EE
GE
