By Jacek Grodecki, director, Geospatial Analysis/Photogrammetric
Engineering, GeoEye (www.geoeye.com),
Dulles, Va.
A clear-cut distinction between satellite and aerial photogrammetric
methodology is gone. In the past, aerial imagery was subjected to
rigorous photogrammetric processing while satellite imagery, being of
only mid to low resolution and typically nadir looking, required much
less sophisticated processing.
Similarly, applications for aerial and satellite images once were
distinctively different. Aerial imagery was collected and processed
mainly for mapping or other applications that require high metric
accuracy, such as digital elevation model (DEM) extraction. Commercial
satellite imagery was confined to remote sensing applications and low
resolution/low-accuracy resource mapping.
Everything changed with Space Imaging’s September 1999 IKONOS satellite
launch, and later with DigitalGlobe’s QuickBird and ORBIMAGE’s OrbView-3
high-resolution satellites, which collect imagery at 0.82 meters, 0.61
meters and 1 meter GSD at nadir, respectively. Similar to aerial
imagery, rigorous photogrammetric processing methods, such as block
adjustment used to solve aerial blocks totaling hundreds or even
thousands of images, are routinely being applied to high-resolution
satellite image blocks. High-resolution satellite image camera models
have been implemented and are supported by most commercial
photogram-metric software vendors such as BAE Systems, PCI Geomatics,
Intergraph Z/I and others.
High-resolution satellite imagery, because of its accessibility to
sophisticated photogrammetric processing methods, photogrammetric
software compatibility, and, most importantly, excellent metric accuracy
characteristics, has been increasingly encroaching on traditional aerial
territories such as mapping and DEM extraction. Aerial imagery, on the
other hand, is being pushed into extremely high-resolution/accuracy
applications, such as high-resolution orthos and civil engineering
projects.
In addition, there has been movement in the opposite direction. With the
advent of wide-area digital aerial sensors with multispectral
capabilities, such as the Leica ADS-40 and Intergraph Z/I DMC, aerial
imagery is being collected for remote sensing applications—an area
previously reserved solely for satellite imagery.
More and more, the choice between satellite and aerial solutions is
being decided on the basis of resolution, attainable accuracy, imagery
accessibility, timeliness and price. Factors such as software
compatibility and rigorousness of photogrammetric processing are no
longer relevant.
Satellite Imagery Benefits
As mentioned previously, one of the decisive factors behind widespread
acceptance of high-resolution satellite imagery by the photogrammetric
user community has been its excellent metric accuracy. This is mostly
attributable to high accuracy of exterior orientation—i.e., satellite
position and orientation—and stability of interior orientation—i.e.,
lens distortion, focal length, detector position on focal plane,
principal point location, etc.
Additional characteristics of high-resolution
satellite imagery may explain its comparative appeal over aerial
imagery. For example, satellite image blocks typically have fewer
images than aerial blocks. This is illustrated by comparing a
digital aerial block with a 1-kilometer footprint against an IKONOS
image block as seen in Figures 1 and 2. Fewer images permits better
radiometric consistency of the final product and simplifies the
block adjustment process.
For IKONOS satellite imagery, and to some extent others, the
Rational Polynomial Coefficient (RPC) sensor model has fewer
adjustable terms than an aerial sensor model, which is a direct
result of the satellite’s excellent relative accuracy (see “RPC
Camera Model Broadens Satellite Imagery Accessibility” at right).
Thus, an image block requires few ground control points (GCPs) to
attain good absolute accuracy.
Because all high-resolution satellite images are directly
georeferenced—i.e., their position and orientation is directly
measured on-board the satellite—satellite images in principle don’t
require ground control. Block adjustment with GCPs, however,
significantly improves the block’s absolute accuracy.
Block adjustment of multiple satellite images
without ground control also improves their absolute accuracy, albeit
to a lesser extent. However, users need to know that the benefits of
ground control propagate uniformly throughout the entire block only
in the case of a stereo image block. This situation is illustrated
in Figures 3 and 4. An image’s horizontal accuracy after the block
adjustment is depicted by its 95 percent confidence region (error
ellipse). A single GCP (symbolized by a triangle) improves accuracy
of all stereo images, although in the case of a mono block it
improves accuracy of only the image onto which it falls. Uniform
accuracy improvement in the case of a mono block is attained only if
all images have at least one GCP.
A more uniform accuracy improvement can be realized by adding a
cross-strip to a mono block as shown in Figures 5 and 6. As before,
horizontal accuracy of an image after the block adjustment is
depicted by its 95 percent confidence region. Adding one GCP to the
block improves accuracy of the image it falls onto and to some
extent of the adjacent image, albeit only in one direction. Adding a
cross-strip to the block results in an almost uniform accuracy
improvement for all images in the block.
As shown in Figure 7, accuracy of orthorectified imagery depends on
the accuracy of a DEM and the magnitude of the elevation angle. The
DEM error’s effect on the ensuing ortho accuracy decreases the
closer the elevation angle is to 90 degrees, i.e., to nadir
direction.
The elevation angles are fixed for a given
aerial camera model, because they are a function of a distance from
the principal point (see Figure 8). Moreover, because typical aerial
cameras have wide fields of view, a majority of image pixels will
have large off-nadir angles (small elevation angles); thus, they
will be significantly affected by DEM errors.
An aerial block typically is collected in stereo, and the DEM used
in the orthorectification process is generated from the stereo
images. Thus, the only way to improve the resulting accuracy of an
orthorectified image is to improve the accuracy of a DEM used in the
orthorectification process—an expensive proposition, because DEM
generation and editing is labor intensive.
High-resolution satellites are highly agile collectors. As shown in
Figure 9, they can pitch and roll as needed to image the target of
interest and collect multiple images on either side of the ground
track. As a result, as shown in Figure 10, the collection elevation
angles can be varied to meet the desired ortho accuracy. In
principle, with the narrow field of view of a typical
high-resolution satellite camera, one could collect all images as
close to nadir as possible to reduce the orthorectification error
due to DEM to essentially zero, albeit at the expense of long
revisit times. In practice, for a given standard product type, the
upper bound of the off-nadir collection angle is determined from the
accuracy of the available DEM, be it external such as the U.S.
Geological Survey National Elevation Dataset (NED) or Shuttle Radar
Topography Mission (SRTM) DEM, or produced internally from stereo
imagery.
All three U.S. high-resolution satellite imagery providers offer a
suite of standard products generated largely automatically in their
ground stations. Some of these products, such as basic or
georectified, which come with a camera model, are intended to be
used as source data for further photogrammetric processing. Others,
namely orthorectified products and DEMs, don’t require any
additional photogrammetric processing by the end user.
Basic images are corrected radiometrically to account for uneven
detector response. The image also is geometrically corrected to
stitch multiple arrays or detectors together so the resulting image
is continuous with no gaps. Images aren’t corrected for geometric
distortions. Thus, the shapes of objects in the resulting images are
severely distorted—e.g., round storage tanks appear as ellipses, and
rectangular buildings appear as parallelograms. Basic images
typically are delivered with the RPC camera model and a simplified
physical camera model. As mentioned, RPCs are compatible with most
commercial photogrammetric software; only some commercial
vendors support the simplified physical camera models, as they’re
different for each sensor.
Georectified images are corrected for
geometric distortions by projecting them onto a reference ellipsoid
and resampling to a standard pixel size in a given map projection,
such as Universal Transverse Mercator. As a result, objects appear
to be distortion free—i.e., round storage tanks remain round, and
rectangular buildings appear as rectangles. Georectified images are
supplied with the RPC camera model, and, like basic images, can be
used as input for further photogrammetric processing, such as block
adjustment, DEM extraction and orthorectification. A sample IKONOS
georectified image is shown in Figure 11.
Individual orthorectified images can be built with external DEMs,
such as NED or SRTM, or with a DEM generated from stereo imagery. If
multiple images cover an area of interest, they are block adjusted
together prior to orthorectification. If available, GCPs also are
used to improve a block’s accuracy. In turn, orthorectified images
are desheared, tonally balanced and mosaicked to produce a seamless
orthomosaic. Figure 12 shows
an IKONOS mosaic before and after the deshearing and tonal balancing
process.
Stereo images can be used as source material
for 3-D feature extraction and DEM generation. Same-pass stereo
images, such as the IKONOS image in Figure 13, are block adjusted
together in the ground station to improve relative orientation and
remove the y-parallax, and resampled to an epipolar projection for
ease of use with 3-D feature extraction software, such as SOCET SET
from BAE Systems. Same-pass stereo collection reduces radiometric
differences between the two images, thus facilitating automatic
feature extraction and DEM generation.
DEMs are generated automatically by the ground station software from
block-adjusted stereo images. DEMs are manually edited later to
improve accuracy. For example, an IKONOS DEM can be produced with an
accuracy of up to 3 meters LE90 at 5 meter post spacing. Lower post
spacing and accuracy DEM products are less expensive to produce,
because they require less manual editing.
Satellite Imagery Applications
As mentioned, photogrammetric applications of high-resolution
satellite and aerial imagery are similar. The only distinction is
that, currently, satellite imagery applications have somewhat lower
resolution and to some extent lower accuracy requirements.
However, satellite imagery offers distinct advantages when compared
with aerial imagery. Primarily, satellite imagery can be collected
anywhere in the world, including areas otherwise inaccessible due to
international borders, conflicts, etc. Image collection also can be
scheduled almost instantaneously, unhampered by the logistical
issues typical of aerial projects. Additionally, high-resolution
satellites can collect multiple adjacent image strips during the
same orbital pass, which means better radiometric consistency of the
resulting orthomosaic and faster delivery to users requiring quick
response. As with aerial imagery, cloud coverage is always an issue.
Examples of photogrammetric applications of high-resolution
satellite imagery include large-area IKONOS orthomosaics, such as
the one over Molokai, Hawaii (top of page). Airfield mapping is
another example of photogrammetric application of satellite imagery.
Figure 14 shows a Terrain Database, Obstacle Database, and Airport
Mapping Database created from same-pass IKONOS stereo imagery for a
U.S. government customer.
Ongoing Development
The trends described in this article are likely to continue. With
the higher resolution and greater collection capacity of the
upcoming high-resolution satellites, such as GeoEye’s OrbView-5 and
DigitalGlobe’s WorldView-1, per-pixel prices of satellite imagery
will decrease significantly. Lower costs, combined with 1.5-foot
image resolution, will further push photogrammetric use of
high-resolution satellite imagery into aerial territory.
