The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
The error equations for parallax ( Ax , Ay ) at point 1 from 
camera A and C is: 
Vi ac (•*■) = AAx ~ ACx + (/ + - (/ + ~J-) A <P C + 
“ A ®c + * y C hK c - ^ac 
v \ac O') = ~ AC y + ~J ± A( Pa ~ A(p c + (/ + y 
2 0) 
-If + y ) A ® C + x a Ak a - x c A K c - Ay AC 
where, Ax ac , Ay AC are the parallax observation values obtained 
by least square matching of images from A and C; AAx , aAy, 
ACx , ACy are the deviation values of the optical centre of A and 
C; x A ,y 4 , x c ,y c are the image coordinates of point 1 projected 
from A and C; f is the principle distance of the cameras; a<p A , 
Aco A , Ak a and a<p c , Aa> c , Ay are the errors of q>, co , k of A 
and C need to be collected. 
The error equations for point 2, 3...12 with other overlapping 
structure from B-C, A-D, B-D and A-B can be derived similarly. 
The following condition equations are accepted 
<Pa + <Pb=° 
0) c + CO D = 0 (2) 
Ka+K b +K c +Kd= 0 
From the equation (1) and (2), 
the A(p A , Aco a , ak 4 , A<p D , Aco D , Ak d can be solved. Thusly, 
the deformed errors due to the light and simple constructed 
mechanical frame can be compensated. The remained error is 
less than 0.4 pixels. This method is important for reducing the 
weight of sensor system making possibility to be accepted by 
UAV mapping. 
Figure 8a, b, c, d and figure 9 show the images taking from 
camera A, B, C, D and the equivalent image constructed by 
above mentioned method. 
4. PHOTRGRAMMETRIC PROCESSING 
4.1 Aerial Triangulation 
According to the feature of data get from the UAV acquisition 
system, a special aerial triangulation program has been 
developed. The advantage features of this software are as 
follow: 1) Making high precision calibration for the geometric 
distortion from normal purpose used digital cameral. 2) Using 
Pos or GPS data combined with image matching to reconstruct 
the topologie relation of the images along the flying direction 
and between the neighbouring lines. 3) All the points in the 
triangulation network are selected and measured fully 
automatically (Figure 10.). 4) multi-view geometric relations of 
the images are solved by large block adjustment with least 
square method (Figurei 1.). 5) The coarse error are detected full 
automatically by large number of redundant observations. 6) 
The result of orientation elements and control points are 
calculated through alternative solution to achieve 1:2000, 
1:1000 or 1:500 scale mapping standard. 
£3 u t MM •■■■■■& 3. 
FigurelO. select and measure the observed points fully 
automatically 
Figureil. multi-view geometric relations of the images 
4.2 DEM Production 
After aerial triangulation the multi-view images are reorganized 
to be divided automatically into basic units as the stereo pairs in 
traditional photogrammetry. Then the DSM is automatically 
generated by image matching and TIN interpolation within 
every unit. It need a little manual interaction operation to 
separate the points upon the building or lie down at grand for 
generation DEM. ALL units are link up to form fully coverage 
DEM (Figure 12.). 
4.3 DOM Production 
The DOM is also produced automatically based on the 
orientation elements and DEM results. Because 80 percent 
overlapping along flying direction have acquired from aerial- 
photography, only the centre part of image in the frame have 
been taken to be rectified into orthophoto imagery(Figurel3.). 
Figure9. the equivalent image of above 4 images