generated in the course of planning of the photographic operation. Such
planning, in turn, is normally accomplished with the aid of the simulator
to be discussed in due course.
The least squares resection generates approximate angular elements of
orientation and revised approximations for the coordinates of the camera.
It also detects and edits out blunders that may have been made in designa-
tion of approximate X,Y,Z coordinates of the points used.
3.4 PRELIMINARY TRIANGULATION
Once all plates have been measured, a sort is executed by the computer
to generate a file designating all comparator measurements for each point.
From this file and from the results of the preliminary resections the reduc-
tion proceeds to perform a preliminary least squares triangulation in order
to generate approximate X,Y,Z coordinates for each point. During this
reduction, automatic editing is again performed to detect blunders which
at this stage mostly result from misidentification of points. During pre-
liminary triangulation refinements to the preliminary corrections for lens
distortion are also made, provided distortion coefficients corresponding to
two distinct distances have been specified. Otherwise such refinement must
depend on the process of self-calibration which can be executed in the
bundle adjustment.
3.5 THE BUNDLE ADJUSTMENT STAR
The separately executed processes of triangulation and resection
employed in the preliminary reduction are executed simultaneously in the
program STAR which is an acronym for Simultaneous Triangulation And Resec-
tion which, in turn, is a description of what is accomplished in a bundle
adjustment. More specifically, STAR is a bundle adjustment with self-
calibration accomodating as many as 6 different cameras and as many as 24
separate exposures (not counting the various exposures merged in certain of
the scenarios for preprocessing of multiple exposures).
In STAR, any control that may be available is exercised with appro-
priate a priori constraints specified by covariance matrices. In situations
involving strong geometrical configurations, STAR may be advantageously
executed without recourse to control (beyond the bare minimum needed to
define the coordinate system to be adopted). Whatever control is exercised,
STAR upon converging will perform an extra iteration dropping all control
and executing a free net adjustment employing the immer adjustment con-
straints developed by Blaha (1972). In the free net adjustment, the coor-
dinate system is so defined as to generate a covariance matrix of the
entire set of triangulated coordinates having a minimum trace (sum of dia-
gonal elements). In this sense, the coordinate system is defined to pro-
duce results of highest possible overall accuracy.
In the self-calibration mode STAR can exercise up to nine error
coefficients for each camera. These consist of any combination of: three
elements of interior orientation Xp,Yp,C; up to four coefficients for
radial distortion K;,K2,K3,Ky4; and two coefficients for decentering dis-
tortion, P;,P,. Each exposure taken by a given camera shares a common set
of error coefficients. When the maximum of 24 exposures from 6 different
cameras is exercised, the reduced system of normal equations is of order
198 (6 x 24 = 144 elements of orientationplus 6 x 9 = 54 error coefficients).
83