the 
ied 
ses 
ect 
are 
| of 
in 
ze) 
The 
| be 
Lis 
ore 
ers 
“to 
1)? 
x ? 
or 
ess 
red 
ase 
ets 
ter 
is, 
get 
sk. 
ion 
een 
ets 
his 
bly 
is 
in 
es. 
ood 
tem 
sen 
For 
ons 
is 
the 
519 
lem 
We 
ill 
ven 
the 
has 
ant 
is, 
key 
nat 
che 
ned 
ith 
|. by 
the 
the cameras.dbf for possible future use. In 
this way, the system updates its knowledge 
just as a human expert does. 
Apart from the body of rules contained in the 
knowledge base, the primary tool used by the 
expert system is the accuracy predictor. The 
accuracy predictor is used with cameras.dbf to 
modify design parameters until the design 
accuracy is met. The accuracy achievable for 
different geometry are shown in Table 1 for 
Wild P31 (f=100mm oc, = 3 um and 2 um); 
Zenzanon etr (f-150mm, c, = 2.4 um) (Chen, 
1985) and Zeiss UMK (f-100mm, o, = 2 ym ). The 
system assumes a safety factor of 1.2, so that 
the design accuracy in this case is #0.092 
(+0.11/1.2). It also assumes that more than 12 
camera stations will be undesirable. 
6. DISCUSSION 
The result of a consultation with the expert 
system is a detailed recommendation which 
describes the equipment and also the initial 
network geometry for the data acquisition 
phase. Note that the choice of camera is user- 
Specified; the system responds by establishing 
if that particular camera is suitable for the 
specified task. In the case reported, we see 
that for the same o, - 2 ym, the P31 requires 
5 stations, while the UMK requires only three. 
The Zenzanon cannot satisfy the specification. 
All these clearly show the importance of 
format size. 
Associated with each recommendation, we can 
develop a cost. Clearly, the cost of a three 
station solution is superior to that of a five 
station solution in terms of time and amount 
of measurements to be performed. The 
interesting problem arises when two different 
systems yield the same number of camera 
stations for a specified objective. By 
formalising the cost calculation, we can 
resolve this dilemma. 
It is interesting to compare equation 4 with 
the coarse object point accuracy indicator of 
Fraser (1989) 
G- = qSo (5) 
C 
as S = D/f. We find that our PEP is a 
formulation of the factor q. Fraser reports 
the value of q varies from 0.5-1.0 for the 
case of strong geometry (four or more camera 
stations). We obtain values of PEF in the 
range 0.6-1.2 for four or more stations 
(Bammeke, 1992b). We suggest that equation 4 
may be taken as a formulation for equation 5. 
7. CONCLUSIONS 
The criteria for verifying the suitability of 
using an expert system approach to solve a 
task are outlined in section 2.3. For the 
cognitive aspects of network design, it can be 
shown that the answer to each of the questions 
{a) to (g) in section 2.3 is 'yes'. Further, 
we have shown that a suitable knowledge-base 
can be constructed using databases, 
appropriate rules and the accuracy predictor. 
We have constructed a prototype expert system 
which can: 
(a) recommend an initial configuration for 
a task. 
(b) ensure the recommended configuration 
satisfies the required accuracy level 
Further work should be aimed at developing a full 
cost model; integrating the expert system with a 
simulator; and increasing the reliability of the 
knowledge base through enriching the knowledge 
content. Finally, the use of the expert system for 
real projects should confirm the validity of our 
conclusions or otherwise! 
REFERENCES 
Bammeke, A. A., 1992a. Development of Mathematical 
Formulae for Predicting Accuracies of Close-Range 
Photogrammetric Networks. Photogrammetric Record 
(in press). 
Bammeke, A. A., 1992b. Designing and Planning of 
Close-Range Photogrammetric Networks: Is an expert 
system approach feasible?. Ph.D thesis, Polytechnic 
of East London (in preparation). 
Chen, L. -C., 1985. A Selection Scheme for Non- 
metric Close-Range Photogrammetric Systems. Ph.D 
thesis, University of Illinois, 216p. 
Fraser, C. S., 1989. Optimization of networks in 
non-topographic photogrammetry. pp. 95-106. In: 
Karara, H. M. (ed.) Non-topographic photogrammetry, 
Second edition. ASPRS, Falls Church, Virginia, USA 
445pp. 
Kenefick, J. F., 1971. Ultra-Precise Analytics. 
Photogrammetric Record 37(11):1167-1187. 
Kretsch, J. F., 1988. Applications of Artificial 
Intelligence to Digital Photogrammetry. Ph.D 
thesis, Purdue University. 
Martin, J., and Oxman, S., 1988. Building Expert 
Systems. Prentice Hall, New Jersey, 455p. 
Ripple, W. J., and Ulshoefer, V. S., 1987. Expert 
Systems and Spatial Data Models for Efficient 
Geographic Data Handling. Photogrammetric 
Engineering and Remote Sensing, 53(10): 1431-1433. 
Sarjakoski, T., 1986. Potential of Expert-System 
Technology for Aerial Triangulation. The 
Photogrammetric Journal of Finland, 10(1):34-46 
Sarjakoski, T. 1988. Artificial Intelligence in 
Photogrammetry. Photogrammetria, 42:245-270 
Shortis, M. R., and Hall, C. J., 1989. Network 
Design Methods for Close-Range Photogrammetry. 
Australian Journal of Geodesy, Photogrammetry and 
Surveying, 50:51-72. 
Shortis, M. R., and Fraser, C. S., 1991. Current 
Trends in Close-Range Optical 3D Measurement for 
Industrial and Engineering Applications. Survey 
Review, 31(242):188-200. 
Xu, Y., 1988. Knowledge Engineering and 
Photogrammetry Tomorrow. International Achieves of 
Photogrammetry and Remote Sensing, 27(B3) :822-830 
ACKNOWLEDGEMENT 
This work has been carried out through the aid of 
a Commonwealth Scholarship tenable at the 
Polytechnic of East London. We also wish to 
acknowledge the advise and assistance of Dr. Peter 
Woolliams, Department of Systems and Computing; and 
Prof. Peter Dale, Department of Land Surveying of 
the Polytechnic of East London.