to represent both residual symmetric radial distortion and the radially 
symmetric component of platen unflatness. Among the coefficients a;3 
through a,s a few moderately high correlations exist (from .85 to .97 for 
photos with a fairly uniform pattern of 25 measured points), but these 
are not sufficiently severe to induce ill-conditioning (even when apriori 
constraints are not exercised). The final coefficients 44950505051 correspond 
to elements of interior orientation and are not normally exercised for reasons 
mentioned earlier. Specific coefficients for decentering distortion have not 
been carried in the above model because of their almost perfect coupling 
with other coefficients. 
While experimentation with error modeling will continue for 
some time, so also will efforts to ascertain the usefulness of calibrations 
established from reductions of photographs taken over densely targeted 
aerial test fields. Such efforts are to be encouraged, for they can lead 
to better understanding of processes generating systematic error. In this 
respect, much more work needs to be done to establish the general signifi- 
cance of anomalous distortion. The isolation of anomalous distortion can 
be greatly facilitated if an ultra-flat reseau platen is employed. 
Investigators employing test fields should adopt a highly critical 
and questioning attitude concerning the adequacy and accuracy of the ground 
survey. Analytical photogrammetry is fast approaching the stage where even 
the best of conventional surveys (of small test fields, in particular) may 
be inadequate. Special measures such as the use of direct and reverse runs 
along orthogonal flight lines should be considered to minimize the effects 
of both random and systematic errors in the control survey. Otherwise, 
emergent systematic effects ascribed to anomalous distortion could actually 
be caused by distortions in the control survey itself and have nothing to do 
with the camera. 
Because the process of self-calibration has turned out to be so 
powerful and is not difficult to incorporate into the bundle adjustment, 
it seems unlikely that self-calibration will be displaced by precalibration. 
Improvements in accuracy by a factor of two to three seem now to be the 
general experience when self-calibration is exercised on blocks with moderate 
to sparse control. On the other hand, improvements are naturally much more 
modest in applications to small and relatively heavily controlled blocks as 
were investigated by Salmenperä, Anderson and Savolainen (1974). 
The significance of self-calibration seems to be that it permits 
the bundle adjustment to approach its theoretical potential. Prior to the 
implementation of self-calibration it had been necessary to exercise an 
excessive amount of control in order to keep the systematic build-up of 
error within acceptable bounds. Now, with the implementation of self- 
calibration, the original promise of the bundle method can be realized — 
very large blocks can be safely and successfully adjusted with only a small 
fraction of the control required by previous standards. This can have 
enormous economic impact on large mapping projects in underdeveloped areas. 
It becomes practical, for instance, in remote and previously unsurveyed 
areas to employ the emerging technology of doppler surveying to establish 
with unprecedented speed, accuracy, and economy an integrated net adequate 
for the control of huge blocks (Brown (1975)). 
As indicated earlier, the concept of self-calibration is not 
limited to the treatment of systematic errors in plate coordinates. As 
originally developed in Brown, Johnson and Davis (1964), it applied equally 
-29-