804
in Los Angeles. The photos are arranged so that the terrain can be
recognized by stereo scopy. Fig. 4 shows the distribution of geo
metric distortion of the uncorrected image.
5. DTM Generation from SPOT Stereo Pairs
For generating DTM from SPOT stereo pairs, it is necessary
to determine precisely the position and attitude of sensor and to
recognize conjugate points. The former is the problem of exterior
orientation and the latter is that of stereo matching.
Exterior orientation can be made by the same method described
in chapter 2 and 3. Since SPOT HRV imagery is sensed with a linear
array sensor, the equation (1) is modified into (8).
x = 0 (8)
y=(I-(N + l)/2)*d
where d : interval of array element
Other notations are the same as chapter 2. The position and the
attitude of sensor can be expressed by monomials and by trinomials
of line number J, respectively.
Although it is necessary for precise orientation to get the
sufficient number of GCPs from the whole area of imagery, plenty of
remarkable points for GCPs may not exist in some areas. The algo
rithm developed by the authors can utilize feature lines like roads,
rivers or shorelines and the data of geodetic observations like
slant distances or levels as control data.
The methods of conventional stereo matching based on image
correlation have the problem of accuracy and reliability. In order
to improve this weakness, our algorithm adopts a triplet matching
method. A conjugate point is searched under the constraint that
three bundles should Intersect at a point as shown Fig. 5.
The result of DTM generation is introduced. The test area is
located on the mountainous area near Mt. Fuji. Fig. 6 shows the
date and off-nadir angles of SPOT imagery. Fig. 7 and Fig. 8 shows
contour image of actual DTM and that of measured DTM, respectively.
The contour intervals of both images are 40m. The accuracy is the
RMSE of 21m and the maximum error of 68m, which is more accurate
and reiiabLe than that by a conventional method.
6. Concluding Remarks
The summaries of the studies introduced in this paper are as
foilows. The image distortions caused by off-nadir viewing can be
corrected by the method based on photograinmetric principles. The
distortions of SAR imagery can be corrected using DTMs. The DTM
generation from SPOT stereo pairs are more accurate and reliable by
the triplet matching method.