le COSS 
the two 
attitude 
g track 
ference 
ig 2(b). 
srences 
is used 
shows 
d along 
ttitude 
olation 
elative 
nd yaw 
| image 
pint,. 
tion of 
(0) 
Q) 
3) 
  
=> ARoll 
  
   
     
Trace of the image 
navigation channel 
patch on main 
channel image 
Real matching 
point at time ty 
Ideal matching 
point at time to 
   
  
   
    
    
  
Main channel image 
Le. 
Along track 
direction 
A 
  
  
  
  
EH EN 
Figure 2(a) Trace of matching points of image navigation 
] Image navigation channel 
channel onto the main image (roll motion only) . 
Trace of the image navigation channel patch on main channel image 
Rus OE M 
  
  
  
  
A Pitch 
Main channel image 
nd 
Actual matching point 
p^ at time tp 
  
  
Ideal matching 
point at timet, 
  
  
  
  
| - 
] Image navigation channel 
ME en patch of data acqusition 
Figure 2(b) Trace of matching points of image navigation 
channel onto the main image (general case). 
189 
Ideal trace line of no atttude change. 
From eq. (3), f(t) are deduced by the inverse 
Fourier transform. In eq. (3), pole will a exist 
when the @Tp is 2n7 -In our data handling 
these pole is set to zero. 
Along with the attitude change between the 
data acquisitions, there is a parallax effect 
caused by the local altitude of terrain witch 
causes an apparent attitude change. Since the 
displacements of two channel CCDs are small, 
the parallax effect is small enough to be 
neglected or corrected by rough order DEM. 
SIMULATION 
In this section, we discuss simulating the 
extraction of the main channel attitude from 
the attitude of the difference in the main channel 
and the image navigation channel. In the 
simulation, the main channel data is random, 
and image navigation channel is generated from 
the main channel data with a suitable time 
delay. The main channel and image navigation 
channel data consist of 8192 points of random 
data. If eq. (3) is divergent, eq (3) is set to zero. 
Figure 4 and Table 2 show the simulated results 
for time delay equivalent of 160, 320 and 640 
pixels along track. In fig. 4 , the ordinate is 
attitude in radians and the abscissa is time. The 
solid and broken lines show the main channel 
attitude and the extracted attitude. The dotted 
line shows the difference in the main channel 
attitude and the extracted attitude. The 
difference between the given and the extracted 
attitudes consists of two components, one of 
which is offset component that doesn't affect 
the attitude correction. The other component 
is residual component which arises error of the 
attitude correction. Table 2 shows the relation 
between the ratio of the average time offsets to 
total data acquisition time and the residual 
component. The residual component decreases 
as the total data acquisition time increases. 
In Fig. 4, the main channel attitude data is not 
a completely periodic function, the bias 
component vibrates. Therefore, the ratio of time 
delay is large, and the attitude extraction is 
inaccurate. In the simulation, the attitude 
extraction accuracy is 6.5x10 6radian when the 
ratio of time delay is 4 %. This is equivalent to 
the 0.52 pixels for images from a satellite 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B1. Vienna 1996