348 PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING, 1976 
EVALUATION OF ACCURACY 
Two methods are at our disposal: We can evaluate accuracy by using check measurements 
and computing from these check measurements the value of adequate accuracy criteria; and 
we can use accuracy predictors, on the condition that they exist and that their reliability has 
been proved. In photogrammetry, where the accuracy problems are very complex, the two 
methods should in my opinion be employed together, and the values supplied from the pre- 
dictor always checked with a minimum of check measurements. 
Now, the methodologic base is, of course, the check measurements. Thus, I desire to insist 
a little upon this problem bearing in mind the photogrammetric area. The principle is to 
compare the results obtained from one particular measuring procedure with the results from 
a "more accurate" measuring procedure. Of course, it is implied that the definitions of the 
measured quantities are the same for the two measuring procedures, which is generally the 
case for precise works of analytical photogrammetry where the measured object is equipped 
with standard targets. 
One of the most frequently used among check measuring procedures is triangulation or 
micro-triangulation. It is generally possible with these methods to determinate a net the 
errors of which are negligible in comparison to the photogrammetric errors. For example, at 
the Institut Geographique National (IGN) in Paris a careful micro-triangulation for a three- 
dimensioned vertical test field (10 m by 12 m by 2 m) has been characterized by the follow- 
ing RMS errors: 
30 um for the x axis (parallel to the test field and horizontal), 
60 um for the y axis (horizontal and perpendicular to the test field), and 
70 um for the z axis (vertical). 
Such an accuracy is enough so as to appreciate spatial photogrammetric errors of magni- 
tude greater than 0.3 to 0.4 mm. I indicate here a very interesting triangulation method*: 
Recording from two (or more) stations the perspective bundles with a theodolite used as a 
camera (no necessity of spirit levels adjustment, as reference to the physical vertical is not 
useful for determination of relative positions of a set of points) associated with the well 
known computational method of relative orientation (scaling from a calibrated tape with 
targets) (Figure 1). It has been shown from experiments that this method is at least as and 
probably more accurate than conventional micro-triangulation for distances between 4 and 8 
meters. Its essential advantage is that it is considerably more simple, rapid, and economic. 
Unhappily it is not always possible to emplace an accurate reference test field, particu- 
larly for very close ranges («2.5 m)!*. However, there is the possibility of considering the 
points in the test field in the photogrammetric adjustment as observed quantities rather than 
fixed; but appreciation of the accuracy of the photogrammetric procedure alone is then 
delicate. 
CRITERIA OF ACCURACY 
The problem is to estimate the accuracy in a given object volume with convenient criteria 
(Figure 2). It is desirable that such criteria have the following properties: Simplicity (com- 
plex criteria indeed could be developed, but would not be practical for daily use by 
engineers); reliability for the whole volume; and, if possible, independent of the object 
Test field 
  
   
  
  
Photogrammetric 
pr e gram 
Calibrated. 
distance 
    
  
  
(Relative orientation- 
Sealing from calibrated 
distances) 
  
  
  
  
  
Ads, 
757 theodolithe 
2™ theo dolithe 
  
  
Fic. 1. Triangulation with a test field and calibrated distance. 
* This method has also been experimented in at the IGN”. I can certify its excellence and con- 
venience.