A EE AO di CE A NL 
  
different positions and distances from a test field of ground 
control points . 
These constraints can be written for two van positions (i) and (j) 
and two cameras (c1) and (c2) as follows: 
«  B!-Bli 
«— REG S REG. Rz (i) » R7 (j) 
(RO G)s RA) (4) 
For the stereo-pair (1), 
  
a, Bi = get x + (yet 2 y92y) 4 (gel - z22y? 
; Red) = RT G* Ry) 
PC gay op Migs e27; à 
° Ra (1) = Ra) Ra (1) (5) 
Where, 
B is the base vector between each steréo-pair 
RB is the rotation matrix between the camera coordinate 
frame (c-frame) and the mapping frame (m-frame)(c) 
Bn is the rotation matrix between the camera coordinate 
frame (c-frame) and the INS b-frame 
R ©?  istherotation matrix between camera (1) and 
cl 
camera (2) 
S-VHS Camera |. 
  
Figure 4. The VISAT System Roof Mount 
98 
3.3 System calibration Results 
A test field of ninety evenly distributed circular targets was 
established for calibrating the system. The coordinates of these 
targets were determined using a network of six points, 
determined by GPS, as base stations for a total station survey. A 
baseline of 1.3 km was used to estimate the azimuth of the 
network using GPS static survey. 
System calibration was performed three times at one month 
intervals. In each calibration test, eight images were grabbed 
from at least five different van positions. For each calibration 
test, the adjustment was calculated twice; with and without the 
relative orientation constraints. Table 5 lists the second 
calibration results for cameras (1) and (2) without the relative 
orientation constraints, where (Ao, AG, Ax ) are the relative 
orientation angles between the cameras and the INS system b- 
frame. Similar results were obtained for calibration (1) and (2). 
The results indicate that the relative orientation parameters are 
different and have an accuracy of + 1 cm for the base vector and 
+ 5 arminutes for the relative orientation angles. The reason for 
having large corrections for the relative orientation angles is the 
fact that the accuracy of aligning the INS is about 3-5 
arcminutes. This adds more distortion to the bundle adjustment, 
which is then reflected in the large corrections to the relative 
orientation parameters. Table 6 summarizes the final results of 
the three different calibrations for camera (1) and camera (2) 
after applying the relative orientation constraints. It is obvious 
that even with one month difference between the three 
calibrations, the changes in the base vector and the relative 
orientation angles are about + 2 mm. and + 2 arcminutes 
respectively. The probable error due to system calibration on a 
distance of 35 m is about 3 cm ( 35 m x tan 2 arcminutes = 3 cm 
) for the relative orientation angles. On the other hands, a 
constant bias of 2 mm in the base between the cameras, 
especially the front ones, can introduce an error of 7 cm, mainly 
along track, for an object 35 m away from the cameras. For more 
details about the contribution of calibration error to the error 
budget see El-Sheimy et. al. (995). 
4. SYSTEM TESTING 
The system was tested in Montreal and Quebec City in 
September 1995. The test areas included open areas, urban 
centers with narrow roads, and minor and major highways with a 
number of overpasses. Some of the Montreal City test results 
will be presented in this paper. They are in six sectors, about 180 
km, surveyed in five days. They were surveyed such that the 
results in all sectors can be used to evaluate the system 
repeatability in forward and backward runs on the same day as 
well as on different days. 
4.1 Relative Accuracy 
The system's relative accuracy can be estimated by the system's 
repeatability and the accuracy of measuring distances. In order to 
test the system repeatability, some well-defined landmarks along 
the test course were used for comparison. Figure 7 shows results 
of a comparison of different runs in both forward and backward 
directions in the same day, taking the forward runs as the 
reference. Figure 8 shows the relative accuracy of the same 
landmarks for day-to-day repeatability, by taking the results of 
one day as a reference. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996 
-. Positional Error (cm) 
Fig 
Positional Error (cm) 
Fig