ON THE REPRESENTATION OF CLOSE-RANGE NETWORK DESIGN
KNOWLEDGE
Scott Mason and Veton Képuska
Institute for Geodesy and Photogrammetry
Swiss Federal Institute of Technology
8093 Zürich, Switzerland
email: mason@p.igp.ethz.ch
Commission V
ABSTRACT:
Achieving a satisfactory network design is a prerequisite to the realisation of high-precision photogrammetric
measurement in industrial applications. Networks are in practice designed by a simulation approach wherein the
expertise of the photogrammetrist is relied upon to overcome the complexity of the task. At the Institute of Geodesy and
Photogrammetry of the Swiss Federal Institute of Technology, a prototype expert system “CONSENS” is being
developed with the aim of testing the suitability of applying conventional AI technology to network design. Two major
steps in building an expert system are to conceptualize and formalize the acquired knowledge. In this latter step,
knowledge is mapped into formal knowledge-engineering representations. Considerations on conceptualizing and
formalizing network design knowledge are described in this paper. Examples from the diagnosis of networks illustrate
the appropriateness of rules and frames in the representation of network design knowledge.
KEY WORDS: Close-Range Photogrammetry, Network Design, Expert System, Knowledge Representation
1 INTRODUCTION
The mensuration potential of optical triangulation tech-
niques is directly linked to the quality of the triangulation
network employed. Careful attention, therefore, must ob-
viously be paid to the design of such networks. The most
practical approach to close-range photogrammetric net-
work design involves a design-by-simulation strategy, an
approach which relies on expertise in order to resolve the
many interrelated and often competing considerations
before a satisfactory design is reached.
A goal of the project Design and Analysis of Spatial Im-
age Sequences at the Institute of Geodesy and Photo-
grammetry, ETH-Zürich, is to examine the feasibility of
applying knowledge-based expert system (ES) technolo-
gy to the task of photogrammetric network design. Be-
cause it is not possible to "prepare meaningful
knowledge representation specifications for a knowl-
edge-based system application in advance" (Walters and
Nielsen, 1988) the methodology being employed to reach
this goal entails development of the ES-based network
design system prototype CONSENS (CONfiguration of
SENSor networks).
CONSENS is comprised of three basic components - an
expert system, a CAD package, and photogrammetric
data reduction (bundle adjustment) software (Mason et
al, 1991). The ES assumes the decision-making role of
the human expert in network design. The CAD compo-
nent provides functionality for representing spatial data
(i.e. surface model of the object to be measured and it's
workspace, and the camera stations of the network) and
446
for performing geometric operations (e.g. point visibility
checking and incidence angle computations) which con-
tribute to the realism of design-by-simulation. The exper-
tise of CONSENS is presently restricted to the design of
networks for the measurement of simple objects, such as
antennae, without workspace restrictions. As ESs should
be applied to narrowly-defined problem domains (Wal-
ters and Nielsen, 1988), development of CONSENS is
focused on automating the network design function with-
in the context of a measurement robot (see Mason and
Képuska, 1992).
The objective of this paper is to present a number of con-
siderations pertinent to the representation of network de-
sign knowledge, as identified from experiences in
building CONSENS. As outlined in Section 2, the deci-
sions on how to represent knowledge in an ES, i.e.
knowledge formalization, are preceded by a step of con-
ceptualisation in which the key attributes of a domain are
made explicit. In Section 3, three different conceptualisa-
tions of the network design task are presented. These
lead to suggestions on the reasoning strategies appropri-
ate to the generic problem-solving processes involved in
this task. Finally, Section 4 reviews the usefulness of two
standard knowledge representations - rules and frames,
in representing the heuristic and structural knowledge in
network design. Examples are taken from network diag-
nosis.
The ES prototyping process is iterative, entailing refine-
ments, re-design and reformulation of the prototype as
the amount and quality of acquired knowledge broadens
