5.5 New calibration of principal point for each frame 
In the third test, a calibration were performed on 
frame one. For each of the frames, including frame 
one again, an outer orientation with the principal 
point as extra unknown were performed. After this, a 
3D calculation with the parameters for each frame 
were done to be able to compare the result with test 
two. The purpose of the test was to find out if it is 
necessary to do a new outer orientation for each 
frame with the principal point as unknown. 
The result shows that the mean deviations from the 
known points are sometimes higher than in test 5.4 It 
seems however that the including of the principal 
point as unknown also lowers the accuracy in some 
cases. 
  
Re-computing the principal point [mm] 
  
Frame d 4 F90 91 
MD MD MD MD 
Pos 1+2 10.7 15.0 10.5 10.8 
  
  
table 7 Re-computing the principal point 
As a comparison, the same test were performed again 
but now with fixed principal point. One would expect 
this to increase the deviations from the known 
points. Instead it actually lowered the numbers. In 
some cases very little, while some were lowered 
considerably. It seems like the small change in the 
principal point is better picked up in a slightly 
different outer orientation than in the parameter 
itself when the control points are limited. 
  
  
  
  
Keeping the principal point fixed [mm] 
  
  
  
Frame 4 FA 90 91 
MD MD MD MD 
Pos 1+2 6.8 7.0 9.8 75 
5.6 Conclusions 
The results from the stability test are in some parts 
easy to interpret while other are more difficult. It 
seems clear that the inner orientation of the camera 
is not completely stable. The principal point cannot 
be defined properly since the digitizing of the film is 
not accurate enough and, at the same time, there are 
no fiducial marks in the camera. The principal point 
may be computed again for each frame if wanted. It is 
however unsure if this improves the result. The 
deviations in the principal point is rather small (up 
to 10 pixels or 60 um) and it seems that a new outer 
orientation with fixed parameters is just as good. The 
other parameters, principal distance, radial distortion 
and scale factor, seem to be rather stable and should 
be possible to use for several sequences. 
216 
6. REFERENCES 
Edgardh, L-À, 1992, Comparison of Precision and 
Reliability of Point Coordinates Using DLT and 
Bundle Approach, Int Arch of Photogrammetry and 
Remote Sensing, vol 9, Comm V. 
Haralick, R.M., Lee, C.N., Ottenberg, K., Nóle, M., 
1991, Analysis and Solutions of the Three Point 
Perspective Pose Estimation Problem, IEEE 
Conference on Computer Vision and Patern 
Recognition, pp 592-598 
Innovativ Vision, TrackEye, Specification of 
Innovativ Vision's Motion Analysis System, Product 
information 
Kállhammer, J-E., 1990. Digitize Your Film Without 
Loosing Resolution. SPIE vo 1358, 19:th International 
Congress on High-Speed Photography and Photonics, 
pp 631-636. 
Larsson, R., 1983, Simultaneous  Photogrammetric 
and Geodetic Adjustment- Algorithms and Data 
Structures, Thesis, Dep. of Photogrammetry, Royal 
Institute of Technology, Sweden