The other solution uses a technique related to the Gauss-Seidel relaxation
method. A submatrix chosen for geometrical reasons, moves in steps across the
original matrix. Once more the escalator matrix of Fig. 16 is seen in Fig. 18
as the shaded area within the heavy contour lines. In a strip flown with 2/3
overlap, five consecutive camera stations are connected by the resulting overlap
of the photographs. Therefore, having six unknowns per station, a 30 x 30
submatrix was chosen, which always contains such a group of unknown orientation
parameters, as they belong to five consecutive camera stations. This partial
System is now displaced along the strip by one station each time, Thus, for a
strip with n photographs one obtains (n-4) such submatrices and consequently
by inversion of these submatrices one obtains five values for each of the
orientation parameters with the exception of the first and last four photographs,
where accordingly fewer values are obtained. The arithmetic means of the roots
of the individual parameters are now computed and considered as the result of
any one iteration cycle. The approximation results thus obtained are used to
continue the computation according to the Gauss-Seidel relaxation method, by
changing the original absolute column, taking into account the coefficients not
in incorporated in the individual 30 x 50 submatrices, together with the values of
il the corresponding orientation parameters, as obtained in the preceding iteration
cycle. The iteration is continued until the roots have converged to a pre-
| established accuracy level. For the economy of the solution, it is of importance
bi . that only in the first iteration cycle the individual submatrices must be in-
ll verted. The roots in the following cycles are then determined by multiplying
the individual inverses by the changed absolute columns.
IX. SUMMARY AND CONCLUSIONS
The analytical solution for the general problem of photogrammetry, as
presented in this report, is not restricted by geometrical or statistical
considerations, because all nine geometrical parameters which characterize a
central perspective can be introduced for any number of photographs. (Compare
[16]). Furthermore, provisions have been made to consider all types of measure-
ments, as they may arise, as erroneous. A least squares treatment results in
the most probable values of the unknowns of the solution, provided that the
residual errors are normally distributed and the various bundles of rays are
generated according to the principle of the central perspective.
| 76
zum "Aum
RETE