158 
4.2 RFM Optimization for QuickBird and SPOT5 Image 
There are two optimization methods for the image RFM models, 
one is the direct optimization method, which needs to 
recalculate the RPC; the other is the indirect optimization 
method, which does not need to change the RPC, but needs to 
carry on the affine transformation in image space, and calculate 
the new image coordinates. 
In the experiments of this paper, the indirect optimization 
method is adopted to optimize two image RFM models. 
Through the affine transformation, some linear system errors 
existing in imaging process could be eliminated, and the RFM 
model of image also can be optimized, then can calculate the 
image pixel coordinates more exactly if the ground 3D 
coordinates are known. 
4.3 Computation Precision of Generalized Stereopair Based 
on RFM 
This experiment selected 30 ground points as CKPs from the 
topographic maps involved in the QuickBird and SPOT5 
images overlaying regions, and completed the construction of 
generalized stereopair and precision checking. According to the 
calculated results, this paper makes the comparison to the 
experimental precision. 
In this experimental comparison includes following three kinds 
of situations: 
(1) the first situation: 
Carries on the optimization to the RFM models of QuickBird 
and SPOT5 image, but not imports the observation value 
weight matrix P in the constructing and calculating of 
generalized stereopair, then evaluates the CKP plane precision 
computed by generalized stereo points under this condition. 
(2) the second situation:: 
Not carries on the optimization to the RFM models of 
QuickBird and SPOT5 image, but imports the observation 
value weight matrix P in the constructing and calculating of 
generalized stereopair, then evaluates the CKP plane precision 
computed by generalized stereo points under this condition. 
(3) the third situation: 
Both carries on the optimization to the RFM models of 
QuickBird and SPOT5 image, and imports the observation 
value weight matrix P in the constructing and calculating of 
generalized stereopair, then evaluates the CKP plane precision 
computed by generalized stereo points under this condition. 
Aiming at these three situations, this paper calculates the plane 
and elevation precisions of CKPs, and the results of calculation 
precisions are as following Table 3 shows: 
have AM but no P no AM but P both AM and P 
max 
+2.018" 
+62.403m 
-1.164" 
-35.991m 
+0.316" 
+9.767m 
Longitu 
de 
min 
-0.011" 
-0.340m 
-0.595" 
-18.398m 
-0.0009" 
-0.027m 
RMSE 
0.717" 
22.185m 
0.825" 
25.500m 
0.108" 
3.348m 
max 
+2.068" 
+63.94 lm 
-0.285" 
-8.824m 
+0.430" 
+13.291m 
Latitud 
e 
min 
+0.0234" 
+0.723m 
-0.0095" 
-0.294m 
+0.0113" 
+0.348m 
RMSE 
0.830" 
25.660m 
0.143" 
4.418m 
0.186" 
5.752m 
max 
+15.403m 
+15.403m 
+8.129m 
+8.129m 
+8.476m 
+8.476m 
Elevati 
on 
min 
-0.093m 
-0.093m 
-0.462m 
-0.462m 
-0.492m 
-0.492m 
RMSE 
4.942m 
4.942m 
2.990m 
2.990m 
2.960m 
2.960m 
max 
2.889" 
89.334m 
1.198" 
37.045m 
0.534" 
16.512m 
Plane 
min 
0.0259" 
0.801m 
0.595" 
18.397m 
0.0113" 
0.348m 
RMSE 
1.097" 
33.921m 
0.837" 
25.891m 
0.215" 
6.665m 
Table 3 comparison of CKPs calculation precision 
5. CONCLUSION 
1 .According to the “no AM but P” results listing Table 2, it can 
be drawn that if not adopt the indirect optimization method for 
the image RFM models, the bias errors of image will cause a 
systemic effect towards the calculation results of plane 
coordinates, while this kind of effect can be eliminated after the 
optimization. 
2. While solving the generalized stereopair using RFM+RFM 
models, the importation of observation value weight matrix can 
effectively improve the solution precision of ground 3D 
coordinates. 
3. According to the RMSE (root mean square error) regulation 
for object points required in the standard (CB/T 17157-1997) 
((Aerial photogrammetry digital mapping standard for 1: 
25000, 1: 50000, 1:100000 topographic map)) , This paper 
constructs the generalized stereopair based on RFM utilizing 
0.6m space resolution QuickBird image and 2.5m space 
resolution SPOT5 image, and obtains the 3D information of 
ground points, then evaluates the precision of generalized 
stereopair. The calculated plane RMSE is 6.665m and the 
elevation RSME is 2.960m. 
Saying from the plane precision, it can absolutely satisfy the 
mapping requirement of 1: 25,000 topographic maps. Saying 
from the elevation precision, it can satisfy the mapping 
requirement of 1: 50,000 topographic maps. 
REFERENCES 
1. (CB/T 17157-1997) Aerial photogrammetry digital mapping 
standard for 1: 25,000, 1: 50,000, 1:100,000 topographic 
map, 1998. the Standard Publisher of China, Beijing.