eight exposures are processed simultaneously to obtain a moderate further im- 
provement in all of the parameters. 
The foregoing set of results demonstrates the effectiveness of the process 
of self-calibration when it is properly implemented. The only external information 
exercised was the minimum required to define uniquely the coordinate system of 
object space (i.e., three translations, three rotations and scale were arbitrarily 
defined). Absolute scale was ultimately derived from measurements of targets 
attached to a calibrated invar tape stretched diagonally across the target array 
(see Figure 8). The general rms accuracy of the triangulation of the 650 targets 
constituting the range turned out to average about .030mm in all three coordinates, 
or about 1:150,000 of the 5 meter length of the sides of the range. À conventional 
survey with first order instruments would, we believe, be hard pressed to match such 
accuracies. 
Over the past decade the Photogrammetric Structural Measurements Service 
of DBA Systems has employed photogrammetric triangulation in over fifty projects 
requiring the precise determination of coordinates of targeted points on parabolic 
antennas ranging in diameter from 1 meter to 100 meters. A general review of this 
continuing endeavor in close-range photogrammetry is provided in Kenefick (1971). 
The process of self-calibration was originally adopted in 1965, with each particular 
project itself providing the observational material needed for calibration. Until 
1968, however, we generally exercised moderately tight constraints on x, , y, , c 
in conformance with values derived from stellar calibrations, due allowance 
being made for the known thickness of special focussing spacers. Since then, our 
policy has been to exercise the process of self-calibration to the fullest extent 
whenever geometrically feasible. This has enabled us to achieve accuracies in 
excess of 1:100, 000 of the maximum diameter of the photographed object from the 
reduction of as few as four to six plates. 
In the usual application of the process of self-calibration, such as is 
illustrated in Table 2, most points appear in all of the photographs; thus, photo- 
graphic overlap amounts to virtually 100 percent over the region of interest. In 
such situations, the coefficient matrix of the normal equations for the adjustment 
is solid, i.e., it is completely filled with non-zero elements. This makes it 
increasingly burdensome to form and solve the normal equations as more and more 
exposure stations are added. Our current computer program for self-calibration 
can accommodate up to nine exposure stations in a simultaneous, in-core reduction 
on a Xerox Sigma 5 computer with 32K memory. This is more than adequate capacity 
to effect an accurate calibration in those projects characterized by total overlap, 
high convergence and diversity of swing angles. However, circumstances may arise 
in which a large number of partially overlapping photographs may provide the most 
appropriate scheme for coverage. For example, large scale coverage of a high dam 
could well entail the exposure of perhaps as many as fifty, systematically overlapping 
frames taken from a fixed pair of exposure stations located on opposite sides of the 
dam. Here, any given point will appear on only a relatively small number of different