B3. Istanbul 2004
id survey, aero
and DEM. 2D (X,
hould be stressed,
of the object point
nd distribution of
of Tie Points
es (or lines), and,
nt features only,
triangulation and
ibsection we only
ts from PRISM
| for extraction of
:M.
tors which locate
in the literature
erators has shown
perform best for
inn and Altrogge,
retical advantages
or (e.g. rotation
racy). We briefly
nd select it in our
y values of square
| directions over a
, we can calculate
low center. Given
N / 2) in an image,
Hndow center are
rk, jen
j+1+D]
+k,j+1)]e
+k,j+1+1)]
| window center,
is more possible
iture point; W is
he W obtains the
take the pixel as
ntrol the density
M nadir image.
nbined matching
on feature point
sy, (b) grid point
square matching
and (d) semi-
| grid matching
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
based on the relaxation matching technique is performed
on a PRISM stereo pair combined with any two viewing
direction of the three images. The important aspect of this
relaxation matching that differs from other area-based
single point matching is its compatible coefficient
function and its smoothness constraint satisfaction
procedure. With the smoothness constraint, poor texture
areas in the image can be bridged assuming the terrain
surface varies smoothly over the image area.
c ) Level 0
Figure 3. Generated pyramid image
Feature matching was firstly conducted on the highest pyramid
level using the collinear equations with the approximate
exterior orientation elements obtained from pre-triangulation
based on a few GCPs. Meanwhile, a window tracking technique
was used to pass the matched candidate points to the lower
pyramid level for fine matching. Figure 4 shows the principle of
our matching approach.
Intermediate
Search o 07
Destination
pyramid level
Figure 4. Principle of window tracking.
To find matches for the feature point, the epipolar geometric
constrain in the forward and backward is used with the
currently valid orientation parameters. All feature points within
the search area are considered to be potential matches for the
feature point in the nadir image. At the same time, we compute
the conjugate point’s ground mapping coordinates using
forward intersection through the combinations of Forward-
Nadir, Forward-Backward, and Nadir-Backward. Only the
combination whose computed ground mapping coordinates are
the nearest is taken as the best match. The matched feature
points will be verified in the next triangulation to check them
intersect one point or not as Figure 5.
Image point
Nadir
image line
Image point /
aget Backward
image line
Image point
Forward
image line
Conjugate
ground
Figure 5. Concept of triplet-image matching for
TLS imagery.
To increase the accuracy of image measurements of feature
points, the least square image matching was used in the final
step of our matching approach. Semi-auto image matching was
used for user to select points in image where fewer points were
matched. The extracted feature points were used as tie points in
next triangulation to get more accurate exterior orientation
correction values or used as seed points to extract more random
points for DEM generation after the triangulation.
2.5 Generalized Bundle Adjustment
A number of research work and applications for the
photogrammetric adjustment of 3-line-imagery have been
conducted (Chen et al., 2001; Lee et al., 2000; Ebner et al.,
1991 and 1992; Fraser and Shao, 1996; Frisch et al., 1998;
Ebner et al., 1999; Kornus et al., 2000). The DGR (Direct Geo-
Reference) method was adopted in our approach since on-board
high precision GPS/IMU data could be used for PRISM
imagery orientation.
In the least square adjustment the GPS/IMU observation values
as the approximate exterior orientation parameters, interior
orientation parameters, ground control information and the
image coordinates of the extracted conjugate points are
considered as observations with corresponding standard
deviations. From this information the unknowns (adjusted
object point coordinates, adjusted offsets of GPS, alignments
and drift errors of IMU) are derived. Additional unknowns are
able to model a more camera motion.
2.6 DEM Generation
The generation of DEM involves the determination of conjugate
points in the three image strips, the computation of object
coordinates for these points and the interpolation of the object
surface. To generating high accuracy DEM further information
