fields came through the use of digital imagery and image
analysis techniques.
Regarding relative orientation, the method in [Schenk et al.,
1991] employs edge detection for selecting matching
candidates which are subsequently precisely matched by
examining their correlation coefficients. Through the use of
image pyramids, matching is fine-tuned and precision is
improved. Reported accuracies in conjugate identification
through this automated method reach 0.2 pixel, which
corresponds to 5-6 um at the finest resolution used in the
reported experiments [Stefanidis et al, 1991]. In a
conceptually similar approach, [Tang & Heipke, 1996] use
geometric coherency constraints and report conjugate point
measurement accuracies on the order of 0.2-0.3 of a pixel,
which at the finer used resolutions correspond to
approximately 3.5 um. In addition to the above mentioned
accuracies, which are comparable to the ones associated with
high-precision analytical photogrammetric processes, the
automated orientation methods offer excellent time
performance (as little as 4 minutes per image pair) and
produce very large numbers of conjugate points per image
pair, outperforming analytical methods in these tasks.
An excellent review of digital matching strategies for point
transfer is given in [Foerstner, 1995]. Automated point
transfer based on graph theory models for the selection of
proper matching combinations has been extended into a
digital aerotriangulation strategy, with accuracies on the
order of 0.3-0.4 pixels [Ackermann & Tsingas, 1994]. In an
alternative approach, multiple image multipoint matching is
used for the automatic determination of exterior orientation
parameters and positioning of selected points within an
aerotriangulation strategy [Agouris, 1992]. The method
proceeds by employing automatic orientation modules for
the automatic generation of approximate photomosaics.
Using these photomosaics, approximate conjugate points
are selected and are subsequently matched precisely by using
a modified least squares based matching solution. Thus, the
technique combines matching and block adjustment,
achieving accuracies on the order of 0.3 pixel at the finer
resolution [Agouris, 1992]. For reasonable resolutions this
can correspond to measuring accuracies of 3-5 um.
3.2 Automated DTM and Orthoimage Generation
DTMs and orthoimages are traditionally two of the most
popular and fundamental geoinformation layers at GIS. Their
automation through digital photogrammetry is having a
major effect in underlining the potential benefits of an
integrated photogeographic information environment.
Automated Digital Terrain Model generation has resulted
from the application of matching techniques to a very large
amount of points in a stereopair of digital images. The
matching solutions define pairs of conjugate points, and
when projected in object space they determine terrain
points. These points are used for the definition of a DTM.
Quite often, various geometric constraints are imposed,
tying together solutions of various points, to ensure
geometric coherence in the object space. The processed
26
points can be selected to form a grid, or be randomly
distributed.
Packages for automated DTM generation from digital
imagery (e.g. MATCH-T) have made the transition from
research into production, becoming available in softcopy
photogrammetric workstations (e.g. Leica/Helava DPW 770,
VirtuoZo). Initial experiments comparing the performance
of such systems to top-of-the-line analytical plotters were
rather satisfactory, with deviations between automatically
and manually determined DTMs being on the order of 0 to 2
m for 1:10000 scale photography and very adverse terrain
(e.g. a glacier area, with abrupt and severe height variations)
[Baltsavias et al., 1996]. Better accuracies can be achieved
for more regular types of terrain. Employing 1:16000 scale
imagery and a DPW system, [Mikhail, 1992] reported DTM
accuracies in the orded of 0.4-1.0 m relative to the heights
derived with state-of-the-art analytical methods. For images
of scale 1:9000 [Walker, 1994] reports accuracies on the
order of 0.009-0.018% of the flying height for DTMs
automatically determined using MATCH-T.
Orthoimages have enjoyed increased popularity with the
transition from analytical to digital photogrammetry, as
they are easy to produce, very suitable for overlay on other
geodata, and can complement, or even substitute,
topographic maps. They convey great amounts of
information and can be considered excellent base maps.
Regarding production, performance and functionality,
digital orthoimage generation has already surpassed its
analog counterpart, and has very successfully made the
transition to production, with numerous firms offering
systems or software modules for orthoimage generation
[Baltsavias, 1993]. Considering both algorithmic and
production aspects, the issue of orthoimage generation is
considered to have reached a satisfactory and rather stable
level.
3.3 Automated Extraction of from
Digital Imagery
Objects
The current state-of-the-art among digital image analysis and
computer vision activities on the subject of automated
object extraction from digital imagery can be found in
[Gruen et al., 1995]. Among the digital photogrammetric
research topics pertinent to GIS integration, this is the one
where a dominant methodological/algorithmic trend aiming
at automation has not yet clearly emerged. Instead, we can
identify numerous approaches and various strategies
developed, often aiming at specific subtasks (e.g.
identifying only roads on high altitude imagery) with
various degrees of success. Among the emerging trends,
monoplotting appears to combine operational ease with the
potential for high accuracy measurements in an operator-
assisted mode [Agouris et al., 1994]. Despite the lack of a
common algorithmic trend, a clear operational trend is
evident: to remove time-consuming and error-prone
measuring tasks from the operator’s duties by performing
them instead through automated modules. The role of an
operator in a modern digital photogrammetric object
extraction process is expected to become limited to guiding
the execution of automated measuring modules, and to
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
provic
which
The a
by
works
gainir
summ
es e & £f? ECD (tj Cr AN A Cet GÀ P9
When
aerotr
to an:
accur:
essen
Thus,
are pe
the ij
simul
inas
overa
From
obvio
integ
data i
integ
categ
integ
4.1
Curre
single
use
photc
to fu
parar
move
poter
and 2
The 1
objec
throu
exist
simp
cons
oper:
