collinearity equations or, in the cases of certain types of
imagery, some other suitable ground to image transformation.
Note that in the DPW there may well be more than two
images, for example four can be displayed in two screen
systems or others may not be displayed at all until called up.
Thus the real-time operations of DPWs and APs are
fundamentally the same, though the former always include
image processing software too, for example contrast
enhancement and filters. Both DPWs and APs, of course, can
vary brightness, the latter by illumination control.
The sections below examine some aspects of DPW software
according to the different applications. It is seen that the main
photogrammetric functions - triangulation, DTMs, orthophotos
and feature extraction - are offered only in the packages from
the traditional photogrammetric suppliers, whereas only a
subset of these functions is available from many others such as
the remote sensing vendors.
3.2 Imagery and mathematical models
All DPWs considered for this paper handle aerial photography.
Many of them also include facilities to treat some satellite
data, usually stereo-SPOT imagery. At the other end of the
equation, there is considerable variety in the ability of the
different DPWs to handle coordinate systems and map
projections. The most flexible include a range of coordinate
systems, datums, ellipsoids and projections, plus the
possibility for the user to program others, whereas more
limited DPWs have a very small range indeed.
3.3 Orientation and triangulation
Naturally, every DPW must include some facilities for
orientation. Interior orientation is mandatory and even the
simplest orthophoto engine, for example, must include a space
resection. Some systems have impressive automated functions,
the interior and relative orientation in the Zeiss PHODIS being
perhaps the best known. Owing to their rather different history,
DPWSs often have orientation philosophies based on bundle
adjustment without the traditional sequential relative and
absolute orientation, though in the Leica-Helava DPWs, a
button has been added so that the classical approach may be
chosen as an option. In many cases, even if complete
automation of the orientation is not there, image matching may
be selected to refine the position of a point which has been
manually selected or measured in one of the images.
The strengths of a DPW are very apparent in the process of
triangulation. Though most vendors have adopted a data
capture approach rather similar to that found in many
analytical plotters, i.e. working along a strip at a time, with
model formation, point transfer, measurement and blunder
checks. The same process on a DPW, however, benefits from
being able to display more than two images at once, to switch
rapidly between them and to recall earlier images for
remeasurement without pain. The user interface in the
Intergraph ImageStation, for example, has proved very
effective (Kólbl, 1996).
Highly automated triangulation is a more recent development
which marks out digital photogrammetry as a step forward. In
principle, the overall configuration of the block may be
derived from GPS data for the exposure stations or from
manual inputs of forward and side overlaps and the offsets
between exposure stations in the individual strips. If each
ground control point is identified and measured manually in at
least one image, then the remainder of the work can be done
automatically, i.e. control points can be transferred, and pass
390
and tie points selected, measured and transferred. The
automated process can include error detection, for example by
model and strip formation, followed by rigorous bundle
adjustment with data snooping to provide a second level of
blunder detection. Several groups have been working
intensively on this, for example Helava, Zeiss, Inpho and Ohio
State University. A summary is given by Fritsch (1996).
Broadly speaking, the groups divide into two schools of
thought. The Helava approach is to define a pattern for tie
points, thereby fixing the image positions in the first image in
which they appear, and then search for the positions in the
other images and measure them by image matching. Failures,
often caused because the forward and sidelaps are incorrectly
defined, are measured manually. The other approaches use
interest operators to find points, which are therefore not
limited in number and may be very dense indeed. Probably the
former approach is more operational at present, the Helava
Automated Triangulation System having been installed by
Leica-Helava on many sites (Miller and Walker, 1996).
Interestingly, at least one user of the Leica-Helava system has
experimented with very large numbers of tie points, making no
attempt manually to measure bad points, i.e. simply omitting
failed points from subsequent stages of the computation.
However unsatisfactory this sounds, it appears to work!
3.4 DTMs
The generation of DTMs by image matching has been a raison
d'étre of digital photogrammetry. This is not the place to
attempt to encapsulate the efforts of a generation of
researchers. It is enough to say that the method is now highly
sophisticated and rather successful, though results are still
unpredicatable in featureless image areas or at discontinuities
such as buildings or trees in large scale photography. Modern
host computers are so powerful that automatic generation of
over 500,000 points per hour is possible, but powerful
interactive editing tools are the key to satisfactory final results.
The two best known techniques are the area based matching of
Leica-Helava, based on Hierarchical Relaxation Correlation
(Helava, 1988; Miller and Devenecia, 1992), and the interest
operator approach of MATCH-T, the Inpho product
subsequently adopted by Zeiss, Intergraph and DAT/EM
(Krzystek, 1991). Matra and ERDAS/Vision International
utilise proprietary techniques and some vendors, such as ISM
and KLT, do not offer products in this area.
Interestingly, the stress on automatically generated DTMs
means that many DPWs are not equipped with the sort of
manual DTM capture functions which are a mainstay of APs.
Without these functions, editing of erroneous elevation data is
a most difficult operation. The subsequent stage of the process
is important too, ie. the use of DTMs for contouring,
generation of triangulated irregular networks (TINs),
computation of profiles and cross-sections, etc. Leica-Helava,
for example, offer KLT's excellent TIN module, whereas for
customers using MicroStation they have opted for the
TerraModeler TIN package from the Finnish company
Terrasolid Oy.
3.5 Orthophotos, mosaics and image maps
Like DTMs, orthophotos are one of the driving forces in the
adoption of digital photogrammetry. Every vendor offers this
facility. There are differences in sophistication, for example
some approaches are completely rigorous in their pixel by
pixel computation, others utilise some kind of anchor point
method to save processing time; resampling algorithms vary
too, but simpler methods like nearest neighbour, bilinear or
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
bicubi
orthop
two v
effects
the bu
DTM.
less in
seems
digitisi
conven
digital
revival
to have
While
photogi
exposu
that in-
While
radiom
nothing
collecti
cases, h
to avoi
buildin;
The add
to form
sensing
quite s;
identifie
Canadia
smartly
software
function
3.6 Fea
In pract
activity
package:
Well on
included
to the n
purchase
KORK :
Thus Le
DAT/EN
Vision c
almost a
Systems
Arguably
as on th
choice o
before. 7
years de
as we Wr
There ha
common
the genei
aesthetic:
popular
packages,
software |
common
the only c
