UNIFORMITY AND PROXIMITY APPLIED TO THE GENERALISATION OF TARGET FIELDS
IN CLOSE RANGE PHOTOGRAMMETRY.
L. C. Anderson (Masters Candidate) and C.J. Bellman (Lecturer)
RMIT University.
Australia.
Commission V, Working Group 1.
KEYWORDS: Design , Close range, Networks, Expert system, Automation.
ABSTRACT:
The design of close range photogrammetric networks can be a difficult task requiring a good understanding of the factors
which influence network design and accuracy. The configuration of the network geometry is a critical factor in determining
the accuracy which can be achieved for a survey. Expert photogrammetrists draw heavily upon heuristic knowledge and
experience throughout this design process. The expert knowledge required has been identified as one of the limiting factors
in the application of close range photogrammetric techniques (Mason, 1994). Expert systems offer a means of automating
the network design process. Mason (1994) proposed a conceptual framework for network design using an expert system.
One of the factors identified in this framework was the need to segment and group the target points into surfaces for which
generic camera configurations are known.
This paper builds on work presented by Mason (1994) and Mason and Kepuska (1991), in particular the investigation of
whether proximity and uniformity are appropriate criteria for the generalisation of target fields into combinations of
planes, cylinders, spheres and cones. Several surface features are reviewed as appropriate indicators of uniformity. The
maximum and minimum curvatures and a function of the surface normal coefficients have been selected as the most
appropriate uniformity indicators for this evaluation of the uniformity and proximity model. Several different
computational procedures which employ uniformity measures to group and/or classify points are reviewed. The paper
details the further development of one of these procedures for the generalisation of target fields using uniformity and
d als BTF. file
ind Tempel, Verlag
1993.
-2,Dümmler Verlag,
anual for Integrated
Bundle Adjustment)
996
proximity.
1. NETWORK DESIGN.
1.1 Network Design For Complicated Objects.
The problems to be addressed in the design of
photogrammetric surveys were identified by Grafarend (1974)
as being four levels of design. This classification of design
problems was also adopted by Fraser (1984) and is as follows:
Zero-Order Design (ZOD) : the datum problem.
First-Order Design (FOD) : the configuration problem.
Second-Order Design (SOD) : the weight problem.
Third-Order Design (TOD) : the densification problem.
The nature of the object, physical constraints of the workspace
and the limitations of available equipment are critical to the
FOD problem and thus the accuracy that can be achieved from
the network. The research presented in this paper relates to the
automation of the FOD process.
The ten constraints and considerations associated with FOD
were dealt with by Mason (1994). These same constraints and
considerations were presented by Fraser (1984, 1989),
however in these earlier articles they were grouped and treated
differently. The ten constraints all limit the placement of
sensors (cameras) within the workspace. Several of these
network design constraints may conflict (Fraser 1992), and the
best compromise is sought when designing an imaging
network.
When designing imaging networks for simple objects a formal
design process may not be necessary. It is often possible for
the photogrammetrist to design an ideal imaging network
simply by viewing the object and its survey site. Design by
inspection however, requires significant skill and knowledge.
For complicated objects, the network design by simulation
process allows for the theoretical precision of object point co-
ordinates to be quantified prior to the actual measurement
taking place and is virtually mandatory for complex objects
(Fraser and Mallison 1992). The simulation process assists the
designer in dealing with the many interrelated and competing
design considerations of an imaging network required for the
survey of a complex object. A limitation of this design by
simulation process is that expertise is generally needed to
efficiently handle challenging cases (Mason 1994). The
requirement for expertise has meant close range analytical
photogrammetry has rarely been applied other than by
experienced photogrammetrists (Mason 1994).
1.2 Expert Systems For Network Design.
Expert systems are computer systems designed to simulate the
problem-solving behaviour of a human who is an expert in a
narrow domain (Denning 1986). The design of strong imaging
networks (FOD) meets the prerequisites of a task suitable for
expert system development (Mason 1994). Expert systems
would play an important role in the development of automated
network design systems (Mason 1994). The advantage of such
an automated system would be to reduce the need for expertise
in close range analytical photogrammetric network design,
apart from the survey of particularly complicated objects.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996