2 = 1/2 1! x (k) * dy^(k)) d ... mean discrepancy of the x-, 
kz3 y-coordinates in position 
Thus a function d = f(c) which gives a connection between the 
principal distance and the mean discrepancy of the points between the 
three models can be defined (Fig. 2). If the results of the three 
combinations are identical, d is 0. In this case the correct value 
for the principal distance has been found. 
  
  
Fig. 2: Definition of the function d = f(e) 
The zero-point, d = 0 = f(c ), can be calculated by using two values, 
Cys C; for the principal distance: C, «QU « boy. The mean 
discrepancy di of c4 gets a negative sign vhile d,? remaíns positive. 
The intersection point c. which has a smaller discrepancy (d4) than c, 
may be used instead of e, now (Fig. 2). This algorithm is to bé 
continued until d; becomes smaller than a threshold for the mean 
discrepancy in position. 
In order to get a good accuracy of the rectified points, it is very 
important that the three images are taken from positions surrounding 
the object. This method cannot be applied if the location of the 
principal point is not identical with the image-center. 
3 INTERIOR ORIENTATION FROM PERPENDICULAR VECTORS AT THE OBJECT 
These algorithms are derived from graphical methods of descriptive 
geometry, where they are used to rectify photographs of buildings. 
Because of the high accuracy of metric cameras, these algorithms have 
been of no importance for analytical photogrammetry. 
3.1 Geometric Elements 
Three vanishing points (F, ,F,,F,) of the edges of a Square block are 
necessary to determine both the principal point and the principal 
distance from one photograph /4/ (Fig. 3). The principal point H can 
be found as the orthocenter of the triangle FT. F,. A cut along I 
and F, gives the length of €; * HO' (0' ... pojection center turned 
into the image plane). 
- 532 —