* Before diagnosing a network using performance
data based on the variance-covariance matrix
(Q,x) of the determined target coordinates as ob-
tained from a bundle adjustment, a pre-diagnosis
Step can be performed. Pre-diagnosis uses the
number of non-parallel rays and statistics on the
convergence angle (e.g. mean and range) between
the rays intersecting at each object point as evalu-
ation data and thereby can provide a quick and
simple means of detecting weaknesses in the im-
aging geometry of the network.
The application of these heuristics in network diagnosis
can be conceptualised in terms of a decision tree, a few
branches of which are shown in Figure 6.It is clear to see
that, with each new decision in the tree, the diagnosis
search space becomes broader.
Are the
number of rays
at each point
O.K.?
Fact: network
satisfies basic
reliability criteria
Fact: some points
have insufficient
rays
11
Is the
convergence
angle at each
point O.K.?
Fact: imaging Fact: imaging
geometry appears geometry weak at
to be strong some points
*ul[
Is the precision
of each point
satisfactory?
Fact: network Fact: some points
satisfies precision fail precision
criteria criteria
*/ul[---
: «ff
Figure 6 Partial decision tree for network diagnosis.
Dashed lines indicate other branches in this
tree.
4 ON REPRESENTING NETWORK DESIGN
KNOWLEDGE
Once the knowledge about a task has been conceptual-
ised, the next step (see Figure 1) in building an ES is to
formalize this knowledge into knowledge engineering
representations. This step is illustrated here by an exam-
ple formalization of the diagnostic task described above.
The application of the two most widely-used knowledge
representations are considered. Firstly, frames are useful
for representing hierarchical knowledge and secondly,
rules are appropriate for representing the heuristic
knowledge in network design. The goal here is not to re-
view the features of these representations as such, but
rather to demonstrate how the representations can be ap-
plied to the knowledge in this domain.
4.1 Example: Representing Hierarchical Knowledge
in Network Diagnosis with Frames
A frame is essentially a structure for holding various
types of knowledge. Conceptually, a frame represents an
item (e.g. a physical object), an idea or hypothesis. The
contents of the frame, called slots, describe that item in
some way (e.g. its characteristics, properties and/or be-
haviour). The chief advantage of having a frame-based
representation is that it provides a means for categorizing
and structuring diverse data-types in the knowledge base,
and a framework whereby not only the data, but also the
structure of the data, can be reasoned with (Walters and
Nielsen, 1988).
The elements of each photogrammetric network can be
categorised into four different classes - camera stations,
images (e.g. photographs), object target points and their
observations, i.e. image points measured in the images.
The physical relationships between instances of these
classes lend themselves naturally to the hierarchical
structuring shown in Figure 7. For instance, the image
exposed at
Observation of
.....
..... Image point k
Figure 7 Hierarchical structuring of configuration data in
network design.
point k is an observation of the object point / and was
measured in the image j. In turn, image j was exposed at
station i. Each network design will be comprised of mul-
tiple stations at which, depending on the SOD, at least
one image will be exposed. Moreover, each object point
will be observed in multiple images; exactly in which is,
of course, an important issue that needs to be addressed
during network design. In addition to camera format,
such factors as point visibility and ray incidence angles
can cause image point “loss” and if not accounted for,
may detrimentally affect the realism of the design simu-
lation" (Shortis and Hall, 1989). In any case, all relation-
FM BD. A | hase AD iat anh” aaa
