O (D m1 (m
"Ta 0 0B -
= © N DO
can be used for checkout while the new model is being developed. The
numerical model also allows debugging problems to be readily addressed,
because questions of whether the problem lies in the photogrammetric model
or in the applications software can be readily resolved by substituting an
existing known good sensor model.
It is also possible to take advantage of the rapid inverse
computation facility by using inverse computations to check intersection
computations, to insure during debug that errors have not been made in the
addressing of the tables, the computation of time, etc. Of course, the
inverse and intersection computation portions of the software need only be
developed once, and then can be used in any applications package desired.
As mentioned above, the inverse computation is generally
considerably more rapid than for a rigorous sensor model. In addition, in
order for an iteration to proceed, partial derivatives of the rigorous
sensor model must be derived. These partial derivatives can be derived
numerically, by repeated computation through the rigorous model with
differential changes to the parameters being made, or analytically, by
taking analytical partial derivatives of all of the rigorous equations.
The first approach is computationally inefficient, and the second approach
is prone to errors in derivation and coding. Neither is as simple as the
numerical approach, which yields identical answers, with fewer iterations
in general, and without any detailed derivation of partial derivatives.
Because the computation of these models is so rapid, they are
very well suited to use in servo controlled systems, or for real-time slew
of digital images in softcopy systems. The microprocessor code developed
will be identical for all sensors, which makes adaptation to ROM very
attractive. Also, timing will be identical for all sensors, which makes
control of the mechanical servo systems much simpler than having to work
the timing problem for each independent sensor.
581
