Each face will be imaged on four spatially separated photographs, and it is 
in this sense that the configuration can be termed multistation. Photogram- 
metric networks of this type, with the camera axes intersecting at good 
angles and the imagery forming a closed spatial net, have inherent geometric 
strength. This is true without any survey control whatsoever and, in 
addition, such set-ups result in a good measure of homogeneity of precision 
in the three co-ordinate directions as will be illustrated later (and has 
been previously demonstrated by Kenefick (1971)). 
This type of multistation configuration has been used on occasions for 
several years. For example Kenefick (1977) has applied a photogrammetric 
network not unlike that shown in Fig 1 for the co-ordinatior of a targetted 
mid-section of a ship. However, the desireability of this type of set-up 
does not seem to be fully appreciated in many areas of potential photogram- 
metric application, especially the co-ordination of engineering and industrial 
structures. There are three main reasons for adopting a multistation solution, 
and all are connected with errors. The first two are concerned with 
Systematic and gross errors (ie essentially with accuracy) whilst the third 
is concerned with the propogation of random errors (ie precision) and will 
be dealt with subsequently due to the need to clarify what precision is 
related to in certain close range situations. 
Systematic and Gross Errors 
  
Accuracy can be considered as the closeness of the adopted solution to the 
"true" solution. Accuracy will, of course, be affected by random errors in 
the individual observations, but the main factor that differentiates 
accuracy from precision is the presence of systematic and gross errors. 
Although most analytical projects require some estimate of precision as well 
the actual object co-ordinate values, this is a meaningless assessment of 
accuracy if significant systematic and gross errors are present. 
The compensation of systematic errors (ie the mismatch between "perfect" 
observations and the mathematical model) has been investigated in recent 
years, and the applicability of the self calibration technique (ie including 
additional parameters in the vector of unknowns) has been extensively 
documented (for example by Brown (1974, 1976), Ebner (1976), Heikkild and 
Inkilà (1978) and Schut (1979)). Once again, however, concepts of self 
calibration pertaining to aerial triangulation must be clearly distinguished 
from close range applications of the technique. In the former, low distortion 
metric cameras are used and the geometric configurations used do not allow 
the evaluation of the three inner orientation parameters. Consequently, the 
self calibration modelling is concerned mainly with film deformations and 
certain residual distortions, and uses essentially empirical polynomial 
forms (Ebner (1976), Grün (1978), Schut (1979)). The possibilities and 
problems in close range studies are far more extensive. Even in instances 
where a relative high precision is required, good non-metric cameras may 
need to be used due to the non-availabilty of suitable metric equivalents. 
In such cases the adequate modelling of systematic errors becomes of para- 
mount importance. In contrast to recent trends in aerial triangulation, 
the inclusion of physically interpretable terms may be more valid (though 
only if likely sources of systematic effects can be readily identified and 
mathematically modelled), but in this case the number of permutations of 
(possibly) significant parameters can become daunting.. In addition there is 
the potential for statistical (rather than physical) correlation of effects. 
For example, we may ask whether glass plates or film is to be used; are lens 
238