209 
particularly for banded or banded-bordered matrices (Ackermann et al 
1973). 
Brown et al (1964), in taking advantage of the characteristics of 
sparseness of the solution matrix, utilized the indirect method of Block 
Successive Over-Relaxation (BSOR).. The unique characteristic of this 
approach lies in the use of an indexing scheme whereby (a) only non-zero 
submatrices of the Normal equations would be formed, and (b) these would 
be stored and operated on in a collapsed form. This procedure, however, 
had two disadvantages, viz., (1) a very slow convergence in case of few 
control points and (2) the problem caused by the impossibility of comput 
ing the inverse matrix. The shortcomings of the BSOR reduction was later 
avoided by utilizing an algorithm called Recursive Partitioning developed 
by Gyer (1967). 
The history of analytical aerotriangulation would not be complete 
if we do not mention the concept contributions on Self-calibration applied 
to aerotriangulation (Ebner 1976, KilpelM 1980) and the inclusion of 
auxiliary data (Inghilleri 1961; Blachut 1957, Brandenberger 1967; Zarzvcki 
1964) or geodetic measurements (Brandenberger 1959, k'ong and Elphingstone 
1972). Any of these would help enhance the quality of block triangulation. 
An important contribution was made by Case (1961) when he showed 
how a substitution of parameters may greatly reduce the rank of the normal 
equation matrix, while at the same time decreasing the error accumulation 
by constraining the parameters in terms of weight or function or both. 
The concept was developed further by Schmid and Schmid (1965) in a general 
procedure for the method of least squares where all elements of the math 
ematical model are considered as observation data, the burden of classifi 
cation being on the weight (constraint) matrix. Thus, for example, by 
simply assigning infinite weight to the control as against the orientation 
parameters, one can use the same computer program for space resection as 
against space intersection. Further contributions have been subsequently 
made (e.g., Mikhail 1970) w T ith regard to constraints. 
Notwithstanding the advancements in analytical aerial triangula 
tion, it has been observed that no result is free from gross errors, search 
and elimination of which are seldom cost-effective. Baarda (1967) devel 
oped for Geodesy a procedure called "Data snooping* 1 which initiated simi 
lar studies in aerial triangulation. Baarda's concep of Internal re 
liability for all bound values of a data snooped system has been used to 
define also External reliability of the upper bounds of the influence of 
non identifiable gross errors (Fdrstner 1985, Gruen 1979). In any case, it 
is also established that a clear subdivision between very small gross 
errors and the random errors may not be possible and, therefore, certain 
errors would remain in the adjusted block. A trend in research toward 
perfection in this regard is being noticed. 
An interesting extension of the self-calibration technique as 
applicable to Bundle Block Adjustment has been tried by Ebner (1976) in 
which additional parameters to consider systematic image deformation (two- 
dimensional) and model deformation (three-dimensional) are included. 
Computationally, these could be handled bv stochastic mathematical models 
(by treating the additional parameters as free unknoxvns) or by treating