In the following sections, the typical application of the 
photogrammetric module is described. 
Im rdinate M men 
The first step of data analysis is the preparation of the data 
files needed for the sensor orientation and as approximations 
for image matching. Among these data files are a list of 
ground control points, image coordinates, camera calibration 
information, and, if available, approximations of the sensor 
positions and attitudes. 
Two windows are used to display a digital stereo image pair 
on the computer screen in a reduced form. This is necessary 
as the digital images very large (6000 x 6000 pixels for SPOT 
imagery and 4000 x 4000 pixels for digitized photographs). A 
small window can be enlarged to full resolution so that the 
operator can identify and measure image coordinates to within 
a pixel, or by using a resampling technique, even to a quarter 
of a pixel. These measurements are done manually on the 
screen. At the same time, the operator identifies control points 
on an existing map. This map must be accurate relative to the 
digital imagery (e.g., USGS quad sheets at 1:24,000 can be 
used for SPOT imagery). Otherwise, ground control points 
must be established by geodetic or terrestrial survey 
techniques. The split screen simulates the functionality of a 
stereo comparator to select points in the left and right images 
of the stereopair. It also allows the user to form a strip of 
photographs by shifting from one stereopair to the next on the 
monitor. If aerial photographs are used, the pixel coordinate 
measurements are transformed into an image coordinate 
system. This coordinate system is defined in the calibration 
report by the camera manufacturer, which is available to the 
user of the aerial photographs. The split screen tool can later 
be used to edit match points in the images. 
The user has an option to check the accuracy of the scanner 
used for digitizing the aerial photographs. The scanner 
calibration program was developed to model the distortions of 
the scanner and to apply corrections to the image coordinates. 
This is an important procedure because many scanners apply a 
second projection to digitize the images, in which case 
additional distortions are unavoidable. These distortions are 
often larger than 3 pixels. 
nsor Orientation 
The sensor orientation is typically done by photogrammetric 
bundle triangulation. This technique is based on collinearity 
equations and is widely used in analytical photogrammetry for 
aerial triangulation. It is a very flexible tool which allows us 
to solve for the exterior orientation parameters of a photograph 
(the exposure station and attitude of the camera), to densify the 
ground points, to derive camera parameters, and to model 
distortions of the sensor system. The bundle triangulation 
technique can not only be used for aerial photographs which 
are modeled by a central perspective geometry, but it can also 
be modified for satellite sensors. Satellites use separate scan 
lines that are acquired sequentially. Each of these scan lines 
has a separate perspective center. Due to its smooth motion in 
orbit, all perspective centers can be lined up along an analytical 
function. Therefore, the geometry of the satellite sensors is 
fundamentally different from aerial photographs, however, it 
can also be modeled by a modified bundle solution. 
Both of these functions were implemented so that the user can 
compute the orientation parameters of either aerial photographs 
or satellite scenes. Furthermore, the user can densify points 
on the ground which were not digitized in the maps network 
densification. As they are positioned very accurately, they can 
be used as a reference in separate images. Sensor orientation 
parameters are important for 3-dimensional positioning of 
points to create the random DEM, and for digital 
orthophotography. 
896 
Image Matching 
The image matching techniques implemented are following the 
area-based approach. To minimize the need of 
approximations, an image pyramid is created by reducing the 
original image four times by a factor of two. At the lowest 
resolution, a square grid is established in the left image and 
matched to the right one. Once this grid has been found, its 
points are projected to the next higher level of the pyramid 
(higher resolution) until the match points finally appear in the 
full resolution image. This first, coarse grid can be edited on 
the split screen monitor so that the user can correct wrong 
matches. These wrong matches typically occur in areas of low 
contrast such as lakes or desert areas. Afterwards, a 
densification is performed to obtain a dense set of reference 
points in the images. For example, in SPOT scenes points can 
be matched as close as every third pixel. 
Two area-based matching techniques are applied: cross 
correlation and least squares matching. Cross correlation 
finds corresponding image points at an accuracy of one pixel. 
It is a very robust technique which locates the points without 
very precise approximations. For the densification, least 
squares matching is used. This technique yields an accuracy 
of a tenth of a pixel in the image. 
Once these sets of corresponding points are available and have 
been edited, an intersection is computed to project the image 
coordinates into object space by virtue of the known 
orientation parameters of the two images. In object space they 
form a random DEM of reference points which can be used to 
interpolate a grid DEM, or they can be connected by triangles 
using a TIN structure. 
DEM Interpolation 
This procedure transforms the spatial random points to a 
regular raster format. It can also filter elevations to eliminate 
wrong matches. The algorithm implemented is based on the 
summation of surfaces, which creates a smooth, analytical 
surface over the whole interpolation area. Wrong matches 
stick out of the surface as spikes. These areas and other noise 
are filtered at this stage to fit the surface to the reference points 
in the best way. The surface of this algorithm also covers 
areas with or without sparse reference points. As a result, we 
obtain the raster DEM that directly forms the elevation layer of 
the GIS. It is georeferenced and can be created in any 
selectable map projection supported by the GIS. 
Digital Orthophoto Generation 
Once a DEM is available, the relief displacement of the original 
images can be corrected by applying the surface elevations to 
this image. Basically, any DEM pixel is projected back into 
the original image where we can resample a gray value. This 
gray value is placed in the corresponding DEM pixel location 
in the orthophoto plane. The relief corrected image is 
georeferenced to the map projection as the DEM. As there is 
no relief displacement, this image can be directly overlaid with 
vector data without showing any off-sets. This is a typical 
problem of regular geocoded images that were not relief 
corrected. 
Although the procedure of creating orthophotos and DEMs 
was described in a sequential form, the digital 
photogrammetric module is very flexible. The user can enter 
the flow chart (figure 1) at any stage if the appropriate 
information is available. The user can also combine data from 
different sources. For example, if the major objective is to 
create highly accurate digital orthophotos at a high resolution, 
they should be produced from large-scale aerial photographs. 
The DEM used to correct relief displacements need not be very 
accurate in this case. Actually, it can be derived from satellite 
imagery, such as SPOT Panchromatic stereopairs. The DEM 
  
wol 
hart 
orth 
SP( 
cori 
botl