The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
1147
Figure 3 shows the geometric stereo model which is used in this
study. The stereo model can be expressed by four parameters,
roll angle (RA), asymmetry angle (AA), convergence angle
(CA) and bisector elevation angle (BIE). The roll angle is the
angle between the epipolar plane and the local vertical, which
represents the rotation of the epipolar plane about flight line
between satellite cameras. The asymmetry angle is the angle
between the bisector of the convergence angle and the
projection of the local vertical onto the epipolar plane. The
convergence angle is the angle in the epipolar plane between
the two rays from the perspective centers of satellite cameras to
the certain ground target point. And the bisector elevation angle
is the angle between the horizontal plane at the target ground
and the bisector ray in the epipolar. In this study, the four
parameters are used together with the B/H ratio to analyze their
relationship to the geometric accuracy of the DSMs and
orthoimages.
4.2 Results and Discussions
22 GCPs measured by VRS-GPS are used as check points to
evaluate the geometric accuracy of DSMs and orthoimages in
vertical and horizontal directions. Table 2 shows the accuracy
results of the DSMs and orthoimages generated from cross
track stereo pairs, and the corresponding stereo acquisition
geometry parameters. Table 3 shows the accuracy results of the
DSMs and orthoimages generated from mixed satellite stereo
pairs, and the corresponding stereo acquisition geometry
parameters. From table 2 and table 3, it can be seen that high
accuracy DSMs and orthoimages can be acquired by using the
Pixel Factory system, and the effect of systematic error is
removed by using 1 GCP.
1112
1113
1114
1214
1314
Q1Q2
1213
RA
15.3
9.7
2.6
1.5
2.3
9.0
6.1
AA
18.8
9.1
12.6
8.1
8.2
6.4
3.8
CA
11.2
35.8
31.8
24.1
9.3
27.7
30.8
BIE
65.9
76.7
77.1
81.8
81.5
78.9
82.8
B/H
0.25
0.67
0.59
0.43
0.14
0.51
0.55
STDV(Dxy)
0.63
0.86
1.04
0.69
1.3
0.59
0.65
RMS(Dxy)
1.7
1.42
1.8
1.46
1.75
1.4
1.18
STDV(Dh)
3.04
1.93
2.34
1.63
8.76
0.8
1.18
RMS(Dh)
3.27
1.96
2.66
2.74
8.77
1.45
1.72
Table 2. Accuracy results of generated DSMs and orthoimages
from cross-track stereo pairs
11Q1
Il 02
I2Q1
I2Q2
I3Q1
1302
I4Q1
I4Q2
RA
11.0
7.1
12.1
2.5
8.8
4.5
4.1
1.1
AA
3.6
18.6
1.2
15.7
5.3
0.7
10.4
3.8
CA
45.8
18.4
34.8
8.6
30.0
23.1
22.7
15.8
BIE
78.4
70.2
77.8
74.1
79.7
85.4
78.8
86.1
B/H
0.88
0.37
0.67
0.16
0.53
0.4
0.44
0.27
STDV(Dxy)
0.79
0.81
0.98
3.46
0.98
0.6
1.14
0.75
RMS(Dxy)
1.43
1.79
1.57
4.76
2.04
1.21
2.05
1.59
RMS(Dh)
2.52
3.88
3.09
30.0
4
4.31
2.86
5.52
6.16
RMS(Dh)
2.73
4.06
3.61
29.7
8
4.88
2.84
5.4
6.07
Table 3. Accuracy results of generated DSMs and orthoimages
from mixed satellite stereo pairs
There are six cross-track stereo pairs in table 2 and 8 mixed
satellite stereo pairs in table 3. In table 2, besides 5 IKONOS
cross-track stereo pairs and 1 QuickBird cross-track stereo pair,
1 in-track IKONOS stereo pair, i.e. 1213, is used to provide a
reference for geometric accuracy comparison.
The geometric accuracy is assessed from horizontal and vertical
direction, respectively. Regarding the horizontal accuracy, it
turns out that there is no significant difference among the in
track IKONOS stereo pair, IKONOS and QuickBird cross-track
stereo pairs, and mixed satellite stereo pairs. For cross-track
stereo pairs, the 1114 shows relatively lower accuracy due to the
difference of image quality caused by the large time interval of
image acquisition. Also, it is confirmed that although the stereo
acquisition geometry has the tendency to influence the accuracy
to some degree, the accuracy does not degrade so much even if
the B/H ratio is small. For mixed satellite stereo pairs, it is
noticed that the accuracy changes considerably with the change
of the stereo acquisition geometry. If the B/H is too small, the
accuracy degrades drastically. Moreover, the influence of the
bisector elevation angle can be confirmed in the case of that the
B/H ratios of two stereo pairs are almost same.
In comparison with the horizontal accuracy results, the stereo
acquisition geometry exerts a remarkable influence on the
vertical accuracy. In the case of cross-track stereo pairs, the
accuracy degrades when the B/H ratio becomes small. In the
case of mixed satellite stereo pairs, since the altitude of
IKONOS and QuickBird satellite are much different, the
geometrical relationship of a satellite stereo pair exerts a critical
influence on the accuracy. Not only the B/H ratio is one of the
main factors that affect the accuracy, the bisector elevation
angle is also an important factor. Moreover, from the accuracy
results, it is confirmed that if the stereo acquisition geometry is
good, the same level accuracy compared with the in-track stereo
pair can also be obtained by both the cross-track and mixed
satellite stereo pairs.
5. CONCLUSIONS
In this study, the in-track, cross-track and mixed satellite stereo
combinations of high resolution IKONOS and QuickBird
satellite images are utilized to generate DSMs and orthoimages.
The geometric accuracy for all combinations is assessed, and
the relationship between the stereo acquisition geometry
parameters and the geometric accuracy is analyzed. It is
confirmed that if the stereo acquisition geometry is relatively
good, the same level accuracy compared with an in-track stereo
pair can be obtained by cross-track and mixed satellite stereo
combinations. It is shown that although the factors such as large
time interval of image acquisition, seasonal change of land
cover, and the difference of image quality have the tendency to
influence the accuracy to some degree, they are not the main
factors. The stereo acquisition geometry of satellite stereo pairs
exerts a critical influence on the geometric accuracy. Also, the
effects on accuracy degradation exhibit different trends in
cross-track stereo pairs and mixed satellite stereo pairs.
Although further analysis is necessary by using more satellite
images and test fields, the assessment results showed the great
potential of using cross-track and mixed satellite stereo images
for spatial information collection.
References
Cain, J., 1989. Stereomodel Acquisition Geometry. Ph.D. thesis,
U.C. Berkeley.