Numerical Photogrammetric Models for High Speed Solutions
Dr. Clifford W. Greve, Craig W. Molander, Dewey Houck
Autometric, Incorporated
Falls Church, Virginia
ABSTRACT
Rigorous evaluation of photogrammetric math models associated
with dynamic sensors is time consuming and computationally intensive. A
numerical model developed by Autometric provides an alternative evaluation
of projective geometry without sacrificing accuracy. Software
implementation of numerical model allows rapid computation of space
intersections and inverses. It also allows "generic" sensor modelling
which minimizes the impact of adding sensors to the system.
INTRODUCTION
Autometric, Inc. has developed a photogrammetric model which
permits the very rapid computation of general photogrammetric projection
equations for any camera system which is at least differentially perspec-
tive. Although originally developed for use in the real-time computation
of servo control parameters within analytical stereoplotters, these
algorithms have proven to be beneficial in general software applications.
Because the same basic implementation of the mathematical model suffices
for all systems, it is possible to build applications software systems in
which all the processing is independent of camera type, once the
appropriate information for the images have been computed and loaded. This
philosophy greatly simplifies the development of software systems, while
yielding accuracies identical to those which would have been rigorously
computed using the theoretical models for the various sensors.
WHAT IS A NUMERICAL MODEL?
A numerical model, as used by Autometric, refers to a
formulation which permits all differentially perspective camera systems to
be processed using the same set of mathematical equations. We define
differentially perspective to mean any camera system wherein a strip of
film or a sensor array is exposed by a lens System, such that all points
along the strip are exposed at a given instant of time. This strip then
becomes a projective framelet of narrow (differential) width, but having
the full projective properties along its width (see Figure 1). In fact,
these projective properties are identical to those of the simple frame
camera system. Since all camera systems strive to present the ground with
some equality of scale between the in-scan and cross-scan directions, the
width of the differential framelet can frequently be expanded to some
finite size without generating any serious errors in geometry. This
approach yields an efficient method for computing the parameters of the
differential framelet and the results obtained by using the framelet are
virtually identical to those obtained with the theoretical model. In fact,
as the width of the framelet becomes differentially narrow, the error
contributed by the frame approximation approaches zero, yielding results
identical to the rigorous model. Autometric has developed an innovative
way to store the data to compute these instantaneous framelet parameters
for any dynamic camera system which is differentially perspective.
572