In such cases, the lack of large residuals does not guarantee the 
absence of gross errors, so that more sophisticated methods are required 
such as those suggested by Grün (1978). By contrast, good multistation 
configurations generally obviate the need for such analyses. The improved 
geometry and increased redundancy means that all but the smallest gross errors 
should be readily detectable from the residuals themselves, because an 
object point imaging on three or more photographs will be associated with 
at least three distinct basal planes (which will intersect at a point). 
Random errors 
Whereas accuuwCct/ is concerned with the discrepancy between the adopted 
solution and the "true" solution, precision relates to the closeness of the 
individual (corrected) observations to the adopted solution via the 
mathematical model. Accuracy is difficult to assess with any certainty, 
whereas the relatively small size (in comparision with aerial triangulation) 
of close range photogrammetric systems allows the computation of all or part 
of the dispersion (variance-covariance) matrix to indicate precision. 
However, we must be careful in assessing what such estimates are relative 
to. With aerial photography, in the final analysis we are relating the 
photogrammetric measurements to a fixed geodetic co-ordinate system, and 
the determination of the control points is considered to be of superior 
quality to the photogrammetric measurements. In close range systems the 
relative precision of control measurements may be more dubious (gross errors 
also being a problem), and in several cases there may be no need to relate 
our results to a fixed co-ordinate system. 
true image position 
    
     
erroneous image position 
erroneous object \ 
position 
¥irue object position 
Fig. 3. On a pair of photographs an error in the image position along an 
epipolar line will not be detected producing an erroneous object position 
in the basal plane.