che 
Ct ali 
>» of 
] the 
ded 
3ts. 
UN 
IHN 
» The 
n. is 
n the 
oints 
e 8 
strial 
many 
n test 
other 
hand, have reduced the measurement effort. Using 60 distances as shown ín 
figure 4 does not change the results. 
4. The accuracy in Z does not change significantly until height differences 
are introduced (test 8). This is probably because the elevation 
differences compared to the flying height is small (nearly flat terrain). 
Table 6 shows the overall accuracy of the different tests, and it may be 
useful to look what happens at the individual object points. The points 
included in table 7 and shown in figure 5 are selected as an example of points 
with constraints in the block. By examining table 7 and figure 5 comparing 
test 2 and 3, it is obvious that the error along the distance direction has 
been removed. For points 68 and 82 the distances are in Y-direction while for 
points 138, 149 and 165 they are in X-direction. The improvement in the 
perpendicular direction or in Z-direction is smaller. When height differences 
are added to the adjustment, the error in Z has almost disappeared. Some 
increase in the errors has taken place in the perpendicular direction but it 
is too small to be corrected by the distances. 
(b) Image Coordinates Contain Affine Film Deformation 
The affine film deformation, introduced into the image coordinates of the 
simulated block, produces a very different error pattern in both the residuals 
and the object coordinates (table 6, tests 9 to 14) from that produced by 
radial lens distortion. The additional constraints have not improved the 
results at all. The main reason is that this type of systematic error does 
not produce significant errors along the coordinate axis that is nearly 
parallel to the distance directions or in Z. Most of the errors in the object 
coordinates are in the perpendicular direction where distances have little 
effect for this size of error. This is clear from table 8, where most of the 
error in points 68 or 82 is in X (distances are in Y direction, see figure 6) 
and in Y-direction for points 138, 149 and 165 (distances are in X- 
direction). 
The overall size of image residuals is very small (less than 1 um), and the 
additional constraints have little effect on them. 
Concluding Remarks 
The effectiveness of the combined adjustment as a tool for error 
detection depends on the following two factors: 
1. Error size. Large errors are very effectively detected by the combined 
adjustment. Points with originally low or no reliability could have a 0.7 
or more redundancy number when two or more distances are measured to these 
points. Systematic errors, due to their small size, could not be detected 
any better, by the residual, using the combined adjustment. However, the 
effect on the adjusted object coordinates (external reliability) has, in 
most cases, been reduced significantly, and thus the overall accuracy of 
the adjusted coordinates has increased. 
2. Error direction. As a rule, terrestrial observations are very effective 
in eliminating the effect of image errors on the adjusted coordinates in 
the direction of the observations. If the observations are distances in 
X-direction, for example, then about 90% of the error in this direction is 
eliminated compared to only 10-35% in the Y-coordinate. The use of height 
differences eliminates virtually all errors in Z. 
Although more detailed studies are still needed, [using other types of 
terrestrial observations at more different configurations | it is safe to say 
155