of 
ry, 
ing 
ive 
ore 
s a 
ion 
ich 
how 
uld 
and 
ist 
rom 
ase 
For 
of 
er 
ral 
orm 
Lin 
of 
orm 
sic 
ts: 
ing 
The 
ied 
ost 
1 in 
set 
ral 
and 
the 
It 
in 
Tt 
ta; 
les 
ied 
the 
ry 
! of 
the 
ce' 
the 
tem 
2.2 Expert system technology and close- 
range photogrammetry 
Recently, the need for photogrammetrists to be 
interested in applying expert system 
technology to photogrammetry has been 
suggested (eg Xu, 1988; Sarjakoski, 1986 and 
1988). Of the four phases involved in 
photogrammetric tasks, suggestions have been 
limited to three: data acquisition, data 
reduction and analysis (eg Sarjakoski, 1988). 
We feel the same concern for the network 
design phase. There has yet been little effort 
in the application of expert system to close- 
range photogrammetry. 
2.3 What kind of problem can be solved 
with expert systems ? 
The verification of the suitability of a task 
for ES support involves finding answers to the 
following questions (eg Martin & Oxman, 1988). 
a) Is the task one in which we can 
express the knowledge required to 
perform the task as a collection of 
facts based on experience ?. 
b) Does the task require an expert to 
carry it out ? Are the specialists 
rare, costly and generally 
unavailable?. 
C) Is the focus of the expertise 
specific, albeit narrow ? 
d) Are experts available for the 
development of the system? 
e) Is the problem insolvable by 
traditional computing methods? 
f) Will the proposed expert system save 
time or money or will it enable a 
task to be performed in a 
substantially better way ? 
g) Does the task occur frequently ? 
If the answer to each of these questions is 
'yes', then there is a prima facie case in 
support of the feasibility of an expert system 
approach to the task. We then need to assess 
whether it is possible to construct a suitable 
knowledge-base. If this is possible, then the 
problem is certainly solvable using expert 
system technology. 
The determination of what constitutes 'the 
knowledge' of the problem is the most 
fundamental aspect in the task of developing 
an expert system (Sarjakoski, 1986, 1988). 
3. DESIGNING CLOSE-RANGE PHOTOGRAMMETRIC 
NETWORKS 
3.1 The problem 
In order to design a CRP network it is 
necessary to make decisions concerning 
a) the camera and initial imaging 
geometry, 
b) the targeting, 
c) the data acquisition scheme, 
d) the measuring instrument, and 
e) the data processing algorithm. 
Some of the above can be expressed purely in 
terms of a 'knowledge' base; for instance, 
choice of camera is dictated by available 
camera types and their parameters can be 
formally described and incorporated within a 
database. The initial imaging geometry depends 
upon such factors as camera type and desired 
survey accuracy and so cannot be described 
455 
using a 'knowledge' base. Nor is there a suitable 
accuracy predictor available without resorting to 
full scale simulation. 
During our research, we soon realised that any 
expert system for network design would have to be 
able to determine if a suggested configuration 
would achieve the accuracy specified for a 
particular task. This led us to develop an accuracy 
predictor (Bammeke, 1992a) which is briefly 
described in section 3.2 and is used to help 
determine initial imaging geometry. All other 
aspects of network design can be handled by the 
expert system itself. As an illustration, we show 
how the problems of targeting and camera choice can 
be handled by the system. These matters are 
discussed in more detail by Bammeke (1992b). 
3.2 Development of Accuracy Predictor 
The determination of the initial approximation of 
imaging geometry is a first-order design problem; 
whereas the selection of camera involves an 
evaluation of the accuracy potential of the camera, 
which is a second-order design problem (Fraser, 
1989). Mathematically we can say that resolving the 
two issues is equivalent to solving for the 
configuration (ie design) matrix A and the weight 
matrix W, in a system.of observation equations of 
the form: 
(A TWA)X - A TWB (1) 
However, we cannot solve this design problem 
without first knowing the accuracy that can be 
achieved with a particular network. Hence there is 
a need to employ mathematical formulae for 
predicting accuracies of close range 
photogrammetric networks. 
Formulae for predicting global estimates of 
accuracies of object-space coordinates in close- 
range photogrammetric multi-station networks have 
been developed. The formula are based upon an 
initial symmetric network geometry consisting of n 
camera stations and certain assumptions (Fig. 2). 
By considering object space point intersection and 
by retaining the scale factor that arises within 
the collinearity equations (Bammeke 1992a) we can 
develop the following formulae: 
  
  
  
1 Spy. ] 
» (cos? (25-1) 25) 
= A 
3 
4, [.2Y2[.DBlg. 
(p: L 
1 =/D 
ag, =~ 2 ri 
n à f 
3 (sin (25-1) 33 
33 2 J 
which give a vector expression 
1 i 1 iip 
Gyr —— t=) 2 =)oy (3) 
(Front) (a] Saar (a) 
where: 
n = No of camera stations in a symmetric network 
$ = angle between consecutive camera axes 
D = average camera-to-object distance,