)w cost
rs. The
Model
tortion.
| in the
with a
apping
e most
and is
some
1 other
and is
or at a
higher
] users
ble for
Jata is
fill the
product
ven at
is lost.
pilation
rces or
source
ay, the
otential
ser has
DTED
2 for a
raction
'ovided
some
vels of
ample,
a lower
needs
| vary it
to edit
' same
curacy
r than
THE PHOTOGRAMMETRIC PROCESS
Photogrammetry has established itself as the main
technique for obtaining precise three dimensional
measurements. It involves the use of overlapping images to
recreate the original stereo geometry of each adjacent pair
of images, from which precise three dimensional
measurements can be derived. Conventional
photogrammetry involves the use of specialist and
expensive plotting equipment to mimic the stereo geometry
at the moment of image exposure using optical trains. The
operator first has to measure calibrated points on the film,
either fiducials or reseau marks, to establish a relationship
between film space coordinates and model space
coordinates. The machines are set up or 'oriented' using a
pair of original hardcopy diapositives in left and right stage
plates. Each stage plate can be positioned with respect to
each other and oriented in x, y and z using threaded
spindles to emulate the precise attitude and position of
each diapositive with respect to each other. In this way any
roll, pitch or yaw in the taking camera or satellite can be
recreated to replicate the attitude and position of each
image at the moment of its exposure. At this point the
images are said to be in relative orientation. Absolute
orientation, based on real world coordinates, requires the
operator to observe and measure known ground control
points in the model space as well.
Once oriented, all residual y-parallax will have been
eliminated, allowing the operator to view the model in
stereo, a projective geometry termed epipolar. When
viewed in stereo, conjugate image points appear in different
positions in each of the images. This 'apparent' movement
of the imaged point is due to the movement of the observer
(in this case the aircraft or the satellite platform) and is
known a parallax. Its measurement forms the basis of
determining height. The only remaining parallax will be in
the x direction, the amount of x-parallax being a function of
height. Using a half mark etched in the optics of each lens,
the operator can ‘float’ the point and move it in a vertical
direction. By placing the point "on the ground" , individual
features in the model can be heighted.
This process has some fundamental drawbacks when
compared to digital techniques. Firstly, it is all based on
very specialised hardware. It is largely mechanical
(analogue) although some plotters can be upgraded to
include linear encoders powered by servo-motors
(analytical) which will drive the operator to pre-defined
points for measuring. Both analytical and analogue
machines however are designed to carry out these single
specific tasks and cannot be used for other applications.
Secondly, the process is a highly skilled one which requires
many hours of training and hence increased staffing costs.
Most of the operations are also very labour intensive,
particularly the collection of height data as each point has
to be visited and measured ‘individually. Experienced
operators can measure anywhere between 6 and 10 points
a minute and like all manual work, it could only be
maintained at the desired accuracy for a specified period of
time, certainly no more than 8 hours maximum. This will
also contribute significantly to overall production costs.
71
AUTOMATED DEM GENERATION
With the advent of sophisticated photogrammetric software
and ever increasing and inexpensive computer power,
softcopy photogrammetric workstations to a large extent
replace the human operator and automatically create the
DTM by means of digital image processing. With production
speeds in excess of 150 points per second, the DTM
production time is significantly reduced. The history of
digital photogrammetry can be traced back to the late
1950's, since which time photogrammetry has undergone a
tremendous change and softcopy photogrammetry now
offers the potential to generate terrain databases with
greater speed, at lower cost and with less training and
photogrammetric skill than ever before.
The major difference between digital and conventional
photogrammetric systems is that images used in digital
systems are in digital format and hence suitable for
processing by computers. If conventional aerial
photographs are used, then they will need to be scanned
prior to input into the system. The systems can also make
use of image data collected digitally, such as satellite
imagery. In this context, the SPOT satellite is the most
commonly used as it currently provides the highest
resolution stereo overlap coverage. However other digital
CCD cameras could also be used.
As with conventional analytical instruments, digital
photogrammetric workstations carry out the same
orientation process in order to model the original stereo
geometry. The principles used are exactly the same, but
the implementation is faster and offers greater ease of use
through intuitive software interfaces. There are a variety of
automated tools based on cross correlation of image
patches to locate and measure fiducials in the image, tie
points, pass points and ground control points. The
correlator can be trained to recognise and measure
fiducials for various camera types and, with the exception
of observing a minimal amount of ground control, the entire
orientation process is automated, requiring very little
attendance and operator time.
The area where most research has been concentrated is
that of automated DEM collection. Sophisticated algorithms
have been developed to replace manual collection and
whilst there are differences between various collection
algorithms, the problem of automating the process of DEM
capture has generally been solved.
The methods that are mostly used are either area-based or
feature-based matching techniques using correlation of
small image templates between image pairs. Once
oriented, the software computes the coefficients of a set of
rational polynomials which summarises the stereo
geometry. These are used by the DEM correlator to
emulate the projective geometry of the cameras. The
normalised cross correlation approach discussed here is an
area based algorithm that digitally correlates points based
on tonal variations present in each image. Areas that have
high tonal and textural variation will be correlated very
quickly as the correlator uses the high frequency
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
