uracy
sible
-1
on of
int, it
ations
ction
errain
> four
1S are
y this
it are
>rrain
ion is
. As
ve in
1, the
led as
uin
DEM
more
y the
s and
ts are
from
e)
set 1 set 2 | set 3 set 4
height error * * * *
range error(avg) 586 [1785 2:27 -0.36
inter. angle(avg) | 14.996 | 14.991 | 14.988 14.988
No. 134 169 247 249
height error xx r* => kk
range error(avg) | -10.43 | -10.42 | -10.42 | -11.13
inter. angle(avg) | 15.008 | 15.009 | 15.009 | 15.009
No. 19174 | 19920 | 20139 | 19481
height error AA IE * 5k xke
range error(avg] | -22.41 | -22.59 | -22.84 | -22.97
inter. angle(avg) | 14.997 | 14.999 | 15.000 | 15.000
No. 530 535 539 519
Table 4: Range errors and intersection angles for all
different DEM height error
* means the DEM height errors are -500~-250m
** means the DEM height errors are -50-50m
*** means the DEM height errors are 250-500m
It is obviously seen that for each data set the range errors
play an important role in determining the height errors.
According to the research, for this pair it is best to
control range error between -8 ~ -12. Also, the chang in
range error is proportional to the image coordinate. If
one pixel is added in the Y direction, the range error is
reduced by 7.9. If one pixel is added in the X direction
the range error is reduced by 0.3. From this
relationship, it is possible to shift the Y2 coordinate of
each matching point. to control all the range errors to be
between -8 ~-12 and see what they will be like. The
results are very encouraging, they do significantly
decrease the DEM height error. The table 5 gives the
DEM accuracy with or without the range errors control
for four different sets, and for each set the DEM accuracy
is reduced at least 50m. The success of the range error
can be explained by the intersection angles. From the
geometric view, for every terrain point, the larger the
intersection angle the more accurate the intersection.
This conclusion is quite consistent here if we inspect the
table 4 again, the intersection angle is the largest one
for the height error -50 - 50m for all the different four
data sets. Changing the range error can also alter the
intersection angle simultaneously. That is, to shift the
Y 2 coordinate based on the range error is actually to
increase the intersection angle and cause very good DEM
result.
set 1 set 2 | set 3 set 4
DEM no control | 89.42 | 90.12| 90.01 | 96.05
DEM control 17.12 | 16.90} 16.81| 17.26
Table 5: DEM accuracy (m) before or after the range
errors control
6. CONCLUSION
From the statistics of the tables in this paper, it is
shown that the CHEOPS is a good approach to stereo
match the SAR. The benefits of the CHEOPS are
analysed. Meanwhile, the function to choose the seed
383
points are also proposed. The best achievement in this
paper is to introduce the idea of range errors which are
very effective to enhance the DEM height accuracy.
Nevertheless, there is still more work to be done,
particularly the performance of CHEOPS on opposite-
side imagery should be analysed and the effectiveness of
the range errors on that pair should be tested as well.
These are all under studies in UCL currently, hopefully,
there will be many excellent results published in the near
future.
REFERENCES
Clark C., 1991. Geocoding and stereoscopy of SAR.
Ph.D. Thesis, University College London
Curlander J.C., 1984. Utilization of spaceborne SAR
data for mapping. IEEE transaction on Geoscience &
Remote Sensing, 22(2):106-112
Day T., Muller J.P., 1989. Digital elevation model
production by stereo-matching SPOT image pairs: a
comparison of algorithm. Image and Vision Computing ,
7(2):95-101
Denos M., 1991. An automatic approach in stereo
matching SEASAT imagery. Proceeding of the British
Machine Vision Conference, 335-339
Denos M., 1992. A pyramidal scheme for stereo
matching SIR-B imagery. International J. of Remote
Sensing, 13(2): 387-392
Dowman I.J., Clark C., Denos M.,1992a. Three
dimensional data from SAR Images. International
Archives of Proceeding of Remote Sensing 29(4): 425-
427
Dowman I.J., Upton M., Knecht J. de, Davison J.,
1992b. Preliminary studies on the application of ERS-1
data to topographic mapping. Proceeding of the First
ERS-1 Symposium ESA SP-359, pp.543-549
Dowman I.J., Chen P-H, Oclochez O.,Saundercock G.,
1993. Height from stereoscopic ERS-1 data. Proceeding
Second ERS-1 Symposium ESA SP-361, pp. 609-614
Leber] F.W., Domik G., Raggam J., Kobrick. M., 19862.
Radar stereomapping techniques and application to SIR-
B. IEEE Transaction on Geoscience & Remote Sensing,
24(4): 473-481
Leberl F.W., Domik G., Raggam J., Cimino J., Kobrick
M., 1986b. Multiple incidence angle SIR-B experiment
over Argentina: stereo-radargrammetric analysis. IEEE
Transaction on Geoscience & Remote Sensing, 24(4):
482-49]
Mercer J. B., 1995. SAR techniques for topographic
mapping . Photogrammetric Week'95 Wichmann. Pp
117-126
Otto GP. ’Chau TK W. 1989. Region growing
algorithm for matching of terrain images. Image and
Vision Computing , 7(2):83-94
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
