ge process. 
wing equa- 
(16) 
calculated 
istortion or 
are known, 
nter /, are 
(17) 
quations. 
(18) 
(19) 
(20) 
(2l) 
d and other 
e: 
Oto 1 
I 
X f4 À 
lute paran 
nded to use 
Kazuo Oda 
  
3. TEST AND RESULTS 
3.1 Test with aerial Photos of Mountainous Area 
The algorithm has been tested with four aerial photos of Tapan Alps area (Figure 2). The test consists of four stages. In 
the first stage, relative orientation parameters between the first pair of image were calculated by observing some match- 
ing point coordinates. Then DSM based orientation was done on the first pair with 7 degrees of freedom under epipolar 
geometry restriction. After that, the orientation parameters of the third image were optimized by performing the algo- 
rithm on the second pair with the parameters of the second image fixed. Similarly the parameters of the 4th image can be 
computed. Digital elevation model of Japan Alps mountain area published by Geographical Survey Institute (50m mesh) 
was used in this test. 
Ist pair 2nd pair 3rd pair 
LL | 1 
2nd image 
  
  
  
  
Ist image 3rd image 4th image 
  
  
  
  
  
  
  
  
  
  
  
DL qu 
] relative orientation 
2.DSM based orientation | 
3. DSM based orientation 
b ie 4. DSM based orientation 
Figure 2. Flow of theTest 
  
  
  
  
  
  
Figure 3 illustrates the process of the optimization of the third pair with 6 degrees of freedom. The second image row of 
the figure shows that difference of the two images is minimized through optimization. Figure 4 shows a mosaic image of 
orthophoto created with resultant orientation parameters and the DEM. 
Before Optimization After Optimization 
  
Mean of Two Images 
  
  
  
  
Difference of Two Images 
Figure 3. Optimization process of DSM Based Orientation 
3.2 Precision of Orientation 
The precision of the algorithm largely depends on resolution of DSM. Figure 5 shows the relationship between DSM and 
precision of orientation. Precision of the orientation is measured by computing mean GCP error at six GCP points. The 
tests resulted that the mean error is almost 1/2 or less than the resolution of DSM. This shows that precision of this 
method is better than resolution of original surface model. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 655