ISPRS Commission III, Vol.34, Part 3A „Photogrammetric Computer Vision“, Graz, 2002 
  
with respect to the reference solution (GPS/INS), then fast 
stereo reconstruction should be feasible. The final, absolute 
image orientation can be performed in post-processing. 
Figure 5 and 6 illustrates typical differences observed 
between the post-processed GPS/INS and free navigation 
mode coordinates, and the corresponding rate of change of 
these differences. Figures 7 and 8 correspond to Figures 5 
and 6, but illustrate the case of a sharp turn, where the 
navigation parameters between the two solutions can vary the 
most. 
  
Ÿ 
Lat, Lea, HE DW qi GE TAMS wu MS 
A T T Fee T 
  
25; ; : : | 
EL. ope o een aet T 
Vu 
ih 
Ans i à 
  
  
i | j i 
Zu — 40 — ub — $ um 
ibriz AIDE - IDEE 
i j ] 
a un AD an mu i 
Figure 5. Difference between post-processed and free 
navigation coordinates. 
Typically, the errors in RO range between 20 arcsec/epoch 
(0.01 to 0.03 m/epoch) for straight portions of the trajectory, 
to 60-70 arcsec/epoch (0.01 to 0.04 m/epoch) for the curves. 
Maximum values observed were ~200 arcsec/epoch and 0.15 
m/epoch for the sharp turns. This amount of error, especially 
in linear offsets, may preclude real-time processing of the 
image pair collected at the curve (which is not really a 
problem from the application’s point of view). In general, 
considering the image overlap of about 1.3 m (~50% overlap, 
see Table 1), the error in relative orientation of about 200 
arcsec (maximum observed in our tests) will translate to an 
~0.9 mm linear offset, which is practically negligible. 
Clearly, the error in the linear component of RO will have 
more impact on the image matching speed and efficiency. 
  
  
  
em = i b 
: | 4; ; | 
1 
TX y UH 
n | necked 
n a 
  
  
  
  
wo mo 
Each bon LIE Te SEB 
Figure 7. Estimated positioning accuracy. 
A - 365 
As already indicated, the navigation solution estimated 
standard deviations are typically at the level of 2-3 cm for 
position coordinates, and ~10 arcsec and 10-20 arcsec for 
attitude and heading components, respectively, based on the 
variance covariance matrix, for short to medium baselines, 
and favorable GPS constellation and signal continuity. Figure 
7 shows typical results for the positions. Naturally, the 
ultimate measure of the georeferencing performance is the 
testing of the integrated (calibrated) system, where the 
navigation and image components work in synch, and 
provide the final coordinates of the features on the ground. 
These, in turn, can be compared to the independently 
acquired ground truth. The MMS discussed here has been 
calibrated and tested using the ground control points 
surveyed by GPS and conventional surveying methods, with 
the accuracy of 1-2 cm in horizontal and 2-3 cm in vertical 
coordinates. The results of these analyses are presented in 
Section 6. 
5. IMAGING SYSTEM CALIBRATION 
The calibration of the imaging components entails two tasks: 
the calibration of the camera system and the establishment of 
its spatial relationship to the navigation system. A target area 
consisting of 10 main ground targets in a 10 m grid and 
extended with satellite points was set up at OSU West 
Campus to support this calibration process. Figure 8 depicts a 
control point with satellite points, having about 1-2 cm 
horizontal and 2-4 cm vertical accuracy, respectively. 
5.1 Camera calibration 
  
Figure 8. Calibration range target points. 
For camera calibration, images were first acquired in a 
laboratory, using a regular grid pattern. Then image 
measurements of all targets from all images were obtained in 
a softcopy environment and subsequently processed with the 
OSU Bundle-Adjustment with Self-Calibration (BSC) 
software. Estimates of the focal length, principal point, and 
lens distortions were then computed. The additional