3IC'M
nodel
ition.
atical
3 the
actual
iy be
rm:
211 be
some
+jXZ
HYz
nstant
SAR
mages
ice to
paper,
entual
h; the
e of
-S-Bus
vare is
permit
x1024
wing.
crystal
assive
bulky,
h less
other
image
rdinate
:xactly
der to
naging
ins the
of the
nas the
tion, it
service
vice is
irdcopy
nd the
ecorded
ches of
digital data in those locations may be obtained to permit the
softcopy aerial triangulation procedure.
The digital data, when it becomes available, is used
to select these small patches surrounding the control and
pass points. The position of these points is measured
precisely on the softcopy display, and the points are also
transferred to all other images on which they appear, using
the softcopy stereo approach. It is important that the stereo
approach be used, since otherwise passpoints must coincide
with identifiable objects in the imagery, which while
desirable may not be practical in all terrain. In difficult
areas, points may have to be selected based upon stereo
fusion of the data, in much the same way that points are
transferred in hard copy triangulation using a stereo point
transfer device.
The measured point coordinates are stored in a data
base which permits later editing should errors be found. The
aerial triangulation software, using banded-bordered
techniques to allow for self calibration of the sensors
involved, performs the solution and presents the results to
the operator in both tabular and graphic presentation.
The Digital Elevation Matrices (DEM’s) are
extracted by automatic correlation, which searches along the
epipolar line for matches. This approach is considerably
faster than simple area correlation, because it uses the
underlying physical phenomena of the taking event to guide
the correlation. The elevation matrix is collected at the
specific posts specified by the user, as opposed to some
systems in which collection is tied to pixel space which
requires a resampling to generate the final grid. If desired,
geomorphological data may be added to improve results in
rugged terrain. The data is edited for blunders using a figure-
of-merit parameter computed during data collection, and
other parameters such as maximum terrain slope. Any
missing points are interpolated from surrounding points
prior to the editing process. For larger scale projects it is
also possible to model the terrain using a Triangulated
Irregular Network (TIN).
The editing process is performed by superimposing
the data over a softcopy stereo model. Extensive editing
tools are available to the operator to permit individual
points to be modified, or entire areas to be modified. This
latter case applies to fitting a plane surface to a parking lot,
enforcing a uniform water level surface on a lake, etc.
After manual editing, all adjacent DEMs are adjusted
together by a least squares procedure, which enforces edge
join constraints between models, and control points derived
during the triangulation process. The output may be
formatted in USGS DEM format, or DMA’s DTED format.
The rectification process projects points
rigorously from the ground plane into the image plane, using
the known position of the pixel in the ground plane and the
DEM. These X,Y,Z values are introduced into the rational
function, yielding the position within the photograph from
which the gray shade should be derived for inserting into the
output orthophoto array. For each output pixel, the corner
coordinates are transformed to the input image, yielding a
125
trapezoid in the input image. Resampling can be done by
nearest neighbor, bilinear, or cubic convolution. Images
are then enhanced to improve the radiometric quality. If film
developing were consistent, and scanning processes perfect,
the process would be routine. However, at this point in time,
there is considerable empirical effort needed to produce
orthophotos which are pleasing to the eye and also maintain
optimum detail. GIS MAGIC™ contains histogram tools
which permit this interactive enhancement to proceed
efficiently.
The geocoded images are mosaicked together along
a series of user-designated line segments. A cursor is
maintained by the system at a corresponding ground point in
each of the two images. Thus, it is very simple for the user
to digitally join two images along any desired cut line. An
automated capability is also available, which provides
digital feathering to smooth the transition along the join
line.
GIS MAGIC™ also has the capability to generate
GIS data. USGS DLG data may be read into the system, and
displayed over orthophotos for determination of the accuracy
of the data. Modifications to those data may be made
interactively, and the resulting data exported in DLG format,
with full topological structuring. Other GIS formats may be
supported.
4. Other Technologies which are being added to
GIS MAGIC™
As we have mentioned, SAIC is constantly
developing advanced technology for our government
customers, and on internal development programs. Some of
the capabilities which we have developed and will be
eventually added to GIS MAGIC™ will now be discussed.
The first such tool is lines of communication
extraction using multispectral imagery. This technique has
proven very successful in extracting lines of
communication, roads, railroads, and waterways, from
Landsat and SPOT imagery. The incorporation of this
capability into GIS MAGIC™ will permit feature extraction
to be performed far more efficiently than at present, and will
reduce operator fatigue considerably. Although the technique
is certainly not perfect, the robust editing capability within
GIS MAGIC™ will permit any erroneous extractions to be
easily corrected.
Another approach which will be added to GIS
MAGIC™ is the resolution sharpening capability being
developed by our Tucson, Arizona office. This approach uses
a hierarchical resolution pyramid coupled with compacted
wavelet compression to permit sharpening of lower
resolution imagery using higher resolution data in one or
more related bands. This offers promise of sharpening
multispectral data such as Landsat using higher resolution
panchromatic imagery at resolutions equivalent to the
National Aerial Photographic Program (NAPP) at a ground
resolution of approximately 27 inches. The approach yields
far superior quality data to that produced by the typical
transformation from RGB to IHS, replacing the intensity
with the higher resolution intensity, and retransforming.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
