2.2 Definition of FEO model
FEO model consists of feature, event and
observation model. Feature is objects such as car,
people and building in the real world. Event is a
spatio-temporal cause, which affects feature.
Observation is the thing that observes event and
feature. Observation data, therefore, is the result
of observed feature and event by observation
(Figure 1).
Figure 1. FEO model
2.3 Feature model
While a wide variety of spatio-temporal data model have been proposed in GIS until now
(discussed in detail in Sekimoto, Shibasaki (1999)), representation with object-oriented
approach has been mainly used in the sense that object-oriented approach can represent the
property of feature intuitively (See, for example, Worboys(1990)). In this flow ISO/TC211,
international standardization activity of geographic information, started in 1994 and has
proposed General Feature Model as conceptual model of geographic information. Spatial
schema and temporal schema, whose higher model is General Feature Model, are defined and
consist of each geometric primitive and topological primitive (ISO/TC211 (1999)).
In this study we follow ISO/TC211 about representation
of features, and also follow ota (1999), extending
ISO/TC211 in terms of including mobile feature. In other
words, feature has
- feature start, end
- feature behavior
- internal change
- effect to the other features (action)
- feature relationship
- feature attribute
and consist of spatial and temporal primitives. We raised
some feature examples of land (immobile, surface),
building (immobile, surface) and car (mobile, point)
(Figure 2).
Though it is necessary to develop conceptual model
with its uncertainty, few works can be seen (For example,
Shibasaki (1994)) and this uncertainty remains to be
studied after this paper.
Figure 2. Feature example
2.4 Event model
Generally, event has been understood in the context of change point in the field of database
(For example, Peuquet (1995)). But considering event as an existence which affects feature,
such as earthquake, traffic accident or development project, event must have spatio-temporal
behavior. This is described by impact model. So event has following components.
- Event start, end
- Event behavior
- internal change
- effect to features (impact model)
- impact method
- impact object
- feature type, relationship, attribute, spatio-temporal range
- Event relationship
- Event attribute
2.5 Observation model
The way of explanation of observational data has an classical but difficult problem. Though
Sinton (1978) claimed that geographic information has three dimensions - time, space and
attribute - and each dimension is fixed, measured or controlled, this remains the problem that
attribute dimension is handled like the other two dimensions. It means that geographic
information, equivalent to observational data in this paper, has some sense like feature object. In
our position observational data is the output result of an attribute of feature measured by
observation controlling time and space, and this control system is sampling model.
An item showing whether observational data is specified to feature or not can reflect the
difference between Euler observation and Lagrange observation. Euler observation measures the
aspects of features by field, and Lagrange observation measures the aspects of specified feature
by flow. For example, measurement only at an intersection or measurement using video camera
can capture the state of unspecified cars. Besides this usual data, measured data by GPS on some
specified car is expected to be integrated. That's why an item showing specified condition is
important. Observation model has
- Observation start, end
- Observation behavior
- internal change
- sampling model
- sampling method
- sampling object
- feature/event type (with an index of specified feature/event or not)
- feature/event relationship, attribute, spatio-temporal range
- sampling result (observational data)
- sampling error (observational error)
- Observation relationship
- Observation attribute
2.6 Discussion
Though it is only natural to need feature model the necessity of event model and observation
model is considered as follows.
Real world consists of a wide variety of features, which have internal changes and changes by
external impact. External impact means that another feature gives some impact on the feature
like car accident, but also means external abstract cause such as an earthquake, climate or
amendment. Of course these can be represented as feature, but it is much easier to handle to
regard these external abstract impact as event when we introduce a control concept like national
land management, especially in GIS which aims to represent and estimate the real world
efficiently and naturally.
This shows that it is closely related to the following matters. Namely in Artificial Intelligence
(AI) and Control Engineering, studies have been done in terms of how to describe the real world
in the computer, considering how the dynamic system like robot or plant adapts to the outside
world. For example, Allen (1984) and Mcdermott (1982) have proposed the existence of event
in order to overcome the poor expression ofsituation calculus in 1970's. Particularly Allen and
Ferguson (1994) classified and defined the kind of
event.
As for observation it helps us to consider that
how the robot regards the observational data. The
explanation, that is, is that the state observed by
observer results in the observational data with
some errors. The state space representation
(Figure 3) in the field of control engineering has
the state space and the external disturbance affects
to the state space, the observer that observes the
state. Thus this shows that state space
representation is appropriate for description of
real world from a viewpoint of control concept above. FEO model, therefore, is conceptual
model and more realistic approach for faithful representation and estimation of dynamical world
in the computer.
Figure 3. Comparison between
state space representation and FEO model
