> quite
of the
lel will
MS-2p
Urther
can he
linates
can be
ters of
initial
lations
image
)0 and
er, the
. Bui,
ultipli-
iber of
object
ns. In
ssed in
linates
rocess.
stereo
g and,
—
Ono Tetsu
finally, stereoscopic image editing. The majority of operational stereo matching algorithms for aerial photog-
raphy require epipolar resampled images. However, conventional epipolar resampling methods are not directly
applicable to satellite line scanner imagery. Proposed approaches for SPOT image resampling have included
approximate methods, but mostly involve a requirement for accurate attitude parameters (Otto, 1988).
Here, again, the 2D affine projection model can be used as the basis for epipolar resampling. This method
is straightforward to apply without accurate attitude parameters. The first author proposed the method of
epipolar resampling using 2D affine projection model, which requires only over 6 stereo matched image points,
a principal distance c and an incidence angle of w (Ono, 1999). Experimental results in the study show that
the accuracy better than half a pixel of residual vertical parallaxes is achievable.
After stereo matching of satellite images is completed, D'TMs are computed from stereo pairs of the image points
coordinates. In case of using standard central perspective model, ground coordinates are easily calculated by
the inverse collinearity equations. 2D affine projection model, however, requires iteration in the calculation of
ground coordinates. The reason for this is that height difference AZ is used in transformation form from affine
imagery to central perspective one (inverted form of Equation 4). But, according to our experience, the number
of the iteration is no more than twice.
After elevation of all grid ground points is obtained by interpolation, the pixel value of the corresponding image
point to each ground point is also acquired by resampling. Ortho-imagery is thus generated as a by-product in
a process of DTMs production.
4 PRACTICAL EVALUATION
In this study, accuracy tests for SPOT ortho-image generation were performed with existing DTMs at Hanshin
area in JAPAN.
4.1 Test Fields and Images
Table 1 shows the data of the test images. The stereo scene covers Hanshin area (Osaka, Kobe and the suburbs)
in JAPAN. The southern area of test field is an almost flat urbanized area. The northern area and the western
area are mountainous. The maximum height difference is about 1,000m. 141 ground points were previously
observed by aerial photogrammetry and GPS. The Estimates of accuracy of these points are 0.3m in planimetry
and 0.6m in height. 9 points of them were used as control points for orientation. The rest points were utilized
for evaluation of generated ortho-image. Image coordinates of the corresponding points to GCPs and check
points were measured manually. The measurement accuracy of these points is probably 1/2 pixel to 1/4 pixel.
Figure 2 shows the test images and the distribution of the measured points.
Table 1: Test Image Data
Left Image | Right Image
Image type SPOT pan Level-1A
Date 1996.11 1995.2
Lat./Long. N34.7/E135.5 | N32.7/E135.2
Incident angle L23.0 R17.9
To serve as height information, elevations were extracted from existing DTMs with about 50m grid spacing,
which are supplied by Japan Geographical Survey Institute. The height resolution of the DTMs is 1m, but the
accuracy are supposed less than 5m. Since spatial resolution of SPOT imagery is about 10m, the DTMs are
interpolated to 10m grid by bi-linear method.
4.2 "Test Results
In order to obtain the orientation parameters of each of left and right SPOT images, orientation of single
image was conducted. Image coordinates are calculated by collinearity equations from the obtained orientation
parameters and ground coordinates of a check point. Accuracy of the orientation was evaluated by RMS errors
Which are derived from difference of image coordinates between an observed image point and the corresponding
check point. The same way was conducted for ascertainment of the ortho-images.
Table 2 shows accuracy of ortho-image and the orientation. According to the results, accuracy of ortho-image
in flight direction is almost same as one of orientation. In the other hand, ortho image has a larger error
than orientation in scanning direction. As expected, errors in the scanning direction were more significantly
influenced by distortions in the DTMs.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 675