action,
theless,
mercial
ms and
evels is
omated
several
ractical
tomatic
mage.
y as we
is third
ntation,
>. These
; than 4
images)
ling the
creasing
scanning
ta to the
nces in
metry is
w years
k RAD
ring and
\ single
to scan
eed and
ical. For
le of 20
MB per
n Square
) images
hen one
compares the size of this tape to a roll film canister, it is very
impressive. The advantage that digital photogrammetry can
derive from this efficient stream of data is growing
continuously. One example of a very efficient production
system is at the National Land Survey of Sweden (Johansson et
al, 1995). Here, disk RAID systems and FDDI high speed
networks permit the efficient manipulation of large quantities of
imagery.
3.0 IMAGE TRIANGULATION
The rapid availability of large blocks of digital imagery has
given rise to one of the most successful areas of automation in
digital photogrammetry - aerial triangulation. Several vendors
and researchers are making good progress in this area (Fritsch,
1995). In the spring of 1995, Leica-Helava introduced the
Helava Automated Triangulation System (HATS). This is an
optional module in SOCET SETG (Softcopy Exploitation
Tools), the standard software suite on the Leica-Helava Digital
Photogrammetric Workstations. HATS uses area based
matching to transfer tie and control points from image to image
in the block, leaving the user only to measure ground control
points (in at least one image) and help the system in the case of
failed points. HATS continues to advance and is proving to be
very productive. It is discussed in detail and with several
practical examples by DeVenecia ef al. (1996) and Miller and
Walker (1996). Typical block triangulation rates are currently
in the neighborhood of 10 minutes per image. This includes the
time for measuring tie and control points, blunder detection and
remeasurement, and simultaneous block solution. Not only is
the productivity impressive, since the above timing must be
compared with the sum of the times for the analogous
operational phases of analytical triangulation using point
marking devices and comparators or analytical plotters, but the
accuracy is excellent also as shown in Table 1 below, which
lists the root mean square errors (rmse) for the ground check
points for some test areas which have been triangulated with
HATS. The GSD represents the ground sample distance of one
pixel.
Test Area rms X rms Y rms Z GSD
Forssa 0.054m 0.057m 0.094m 0.12 m
28 images
WiscDOT 0.085ft 0.089 ft 0.240 ft 0.15
60 images
Table 1. Results from HATS
Typical internal image RMS errors are 0.3 of a pixel. Table 1
reflects the error at check points which may be inflated due to
additional point marking or identification errors. The following
paragraphs summarize the current state of image triangulation
automation.
3.1 GPS and Aerial Triangulation
In HATS, automation can start with GPS inputs. Several of
today's sensor/camera systems offer GPS ready cameras
including the Leica RC30 with ASCOT, which provides flight
planning, navigation and pinpoint photography. ASCOT
provides in-flight GPS coordinates, but Leica also provide SKI
software for post-processing. The resulting GPS file of camera
Positions can be associated with the scanned images. This
251
immediately "sets up" the image block for automated or semi-
automated measurement of control and tie points. This lessens
the number of inputs the user must perform and of course
reduces mistakes. GPS can reduce the number of control points
required by the block to attain a given level of accuracy. GPS is
also generally more accurate than estimates provided by the
user and this gives rise to faster automatic measurement as well
as more reliable measurement. The success rate of automatic tie
point matching as well as driving the user to control points is
enhanced. This leads to substantial productivity gains as well as
greater accuracy and reliability. It is somewhat mundane
automation but it is very useful and impressive to be able to
drive to any control point in any image of a 400 image block in
less than one second.
3.2 Point Measurement
Perhaps the largest gain in productivity for triangulation is
through automatic point measurement. As part of HATS, the
user can choose to execute the automatic point measurement
process. This process will measure tie points throughout a block
of images and it will transfer measured control points to other
overlapping images. This process can be executed on aerial
images as well as satellite images such as SPOT and JERS.
Typical measurement time for this process is in the range of 1
to 2 minutes per image. This compares very favorably against
manual methods. This is particularly true when point marking
is considered. Of coarse, in a digital environment, point
marking is not desirable. The automatic point measurement of
HATS can also exploit a given DTM to better estimate the Z
value in rugged terrain. This permits even higher success rates
for automatic measuement. One of the most startling reasons
for this process being so successful is the reduction in fatigue
for the user. The majority of the points do not need to be
meticulously measured by the human. Instead, the user can be
working on other necessary endeavors while the autonomous
process measures the points for the entire block of images.
This process is typically followed by a semi-automatic process
that drives the user to each missing point measurement. The
software automatically positions the images near the desired
point location and automatically "rectifies" the images for
stereo viewing along epipolar lines regardless of the flight or
scan direction of the images and overlapping strips. This
automatic rectification will also scale the images to each other
so that measuring tie points in images of differing scales is
straightforward. This is particularly useful and necessary when
dealing with satellite image blocks or aerial blocks with cross
strips. It is also very useful when connecting images that
wander along transportation corridors. During interactive
measurement, the user merely positions the cursor near the
desired point and the automatic measurement button can be
used to match corresponding points. Thus even the failed points
can be measured without much fatigue and at faster rates than a
fully manual measurement.
3.3 Blunder Detection
Although blunder detection phases for triangulation have been
around for quite some time, today's digital photogrammetry
provides for several time reducing steps. Blunder detection in
HATS performs image to image relative orientation, model to
model connections, strip to strip connections, absolute
orientation to ground and simultaneous block adjustments. At
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
