The imagery used was a single stereopair of colour
1:25000 scale photographs of an area of Manchester,
UK, scanned at a resolution of 1016 dpi (22.5 microns).
The model contained a mix of topography, including
undulating terrain, urban features, water bodies and
forested areas. The scanned images were utilised on
both the digital systems with the parent diapositives being
set up in parallel on the Zeiss P3 analytical plotter. On the
digital systems the basic photogrammetric procedures
were performed, from basic 'setting-up' of the imagery to
digital terrain modelling and orthoimage generation.
With respect to the orientation procedures, all three
systems (the two digital and the one analytical) produced
comparable results. It must be said that the control points
used were coordinated terrain features, clearly visible on
both images of the stereomodel. Experience has now
shown that problems occur when using imagery where
control points only occur on one of the two images i.e.
mono pug marked control. This makes it difficult to
perform relative orientation monocularly, particularly with
the ERDAS system, where only mono measurement is
possible. With the ERDAS system, control points, used in
the triangulation, must be identifiable on both images.
With the ImageStation it is possible to perform the
procedure in either mono or stereo. The procedure is
quicker and easier in mono rather than in stereo.
Investigation from just visual inspection (through
stereosuperimposition of the DEM on the stereomodel)
identified that the automatic DEM generations were
tremendously variant with respect to the viewed
stereomodel. The consistency of the results varied
according to the terrain being viewed/analysed, with the
greatest consistency appearing to be achieved over rural
areas, compared with urban environments.
Different parameter settings varied the standard of DEM
produced drastically. Notable problem areas included
areas of homogenous texture, shadows, dark slopes and
water features. Urban environments proved particularly
difficult to model, whether attempting to obtain just ground
heights or just roof heights. This did however improve
with the generation of grids with minimal spacing. This
lead to processing times being increased, but individual
buildings were modelled reasonably well, rather than
possibly being completely missed, which occurred with
previous spacings. It was concluded that the photo scale
of 1:25000 was too small to allow reasonable modelling of
the urban features. Clearly some form of statistical
measure is required to analyse the DEMs produced,
rather than only visually checking them against the
stereomodel.
With the use of digital images, the photogrammetrist is
forced into recognising the features which will affect the
accuracies of the output products. The geometric and
radiometric characteristics of the image must be
equivalent to those of the traditional hardcopy
photography. Output may be affected by such things as
scanning resolution, whether the imagery is colour or
monochrome, image compression, image resampling or
various other filtering techniques which are now available
with the movement into image processing.
The image file sizes involved are large. A colour image
(whole photograph) scanned at 10 microns occupies in
the order of 1.7 Gbytes whereas one scanned at 20
microns requires approximately 432 Mbytes. This issue is
obviously significant when large projects are
encountered. Experiences with back-up procedures have
indicated errors with tapes and significant variability in the
time taken to back-off or restore data, especially over
busy networks.
4. CURRENT RESEARCH
4.1 Input Data
Continued experimentation is focused on the accuracy of
the automatic DEM. Two new stereomodels have been
selected with larger scales than previously used and with
a wide range of topographic characteristics. It is important
to analyse the merits and limitations of these algorithms
with respect to a diverse collection of physical
phenomena. The imagery includes a pair of black and
white photographs at 1:3000 scale containing a mixture of
residential and industrial features which is used for
analysis of results in an urban area. The second pair of
images are from colour 1:10000 scale. These are used
for parallel testing within a rural environment. The terrain
encountered has substantial topographical variations with
sudden changes of slope, areas of homogenous
colour/texture and a scattering of rural buildings.
Both sets of imagery were initially scanned at 12.5 micron
resolution, but problems occurred with the use of mono
control, as previously described. For consistency, both
sets of imagery were pugged in stereo and rescanned.
The black and white imagery being scanned at 10 micron
resolution and the colour imagery being scanned at 20
micron resolution. This was performed on NRSC's
recently purchased XLVision OrthoVision 950r digital
scanner. This produced file sizes for each image of 576
Mb for the black and white imagery and 432 Mb for the
colour imagery. Clearly, the storage required for only two
images is large. If a number of models are utilised in a
project then careful planning is required. A feature of
future work will be the use of a run of overlapping imagery
to introduce the testing of DEM consistency over a
number of models.
4.2 Parameter Variations
With both digital systems there are a large number of
variable parameters, for which the user must select a
value prior to DEM generation. It is important for system
users, whether they be trained photogrammetrists Or
scientists grasping the benefits of this new technology, to
appreciate the tremendous variations in output available
due to altering these parameters. This 'black-box
technology must be treated with caution before
acclaiming the benefits of the output.
With OrthoMAX and SoftPlotter there are 16 and 17
variable parameters respectively, most of which accept a
range of values. With the ImageStation, the automatic
358
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
Th
inv
res
aut
coii
ger
spe
acc
the
heic
whi
Wor
but
The
instr
ERC
Imag
split
imag
varie
The
prog
an ir
rathe
6.1 E
6.1.1
The
prodt
requi
orient
fit noi
grid |
bounc
comp
These
prese
effect.
reque: