Next: Implementation details for the
Up: A Software Architecture for
Previous: Previous work
The proposed architecture
According to fig.1,
the desired system
should possess the following capabilities:
- tracking objects of interest
within a single observer by an active camera;
- integration of the data obtained from several
observers to the common scene representation,
by assuming different observer performances;
- coordination of the viewing directions of the observers for
the purpose of achieving a good tracking of the state in the scene;
- robustness with respect to the removal of existing
or adding new observers;
- soft real time performance.
The architecture design is mostly determined by the requirement
that a computer vision algorithm is required to operate in
the real time environment.
Because of the complexity of vision algorithms,
it is favourable to ensure that each observer agent
gets most of the time of a dedicated processor,
and to assign data integration and coordination tasks
to the coordinator agent running on a separate computer.
The resulting architecture is outlined in fig.2:
observers send measurements to the coordinator,
which integrates the data into the common view
and controls the observers behaviour
in an opportunistic manner.
Figure 2:
The top level view of the multiagent architecture.
|
The system organized after fig.2 satisfies the agenthood test
cited in the Introduction,
if the coordinator is viewed as a part of communication infrastructure.
Whenever a new observer registers with the coordinator,
the responsibilities of all observers are rescheduled
in order to obtain a better coverage of the scene.
Next: Implementation details for the
Up: A Software Architecture for
Previous: Previous work
Sinisa Segvic
2003-02-25