Next: Single Object Detection and
Up: Technological Research Issues
Previous: Appearance Sphere Camera
To realize the real time object motion understanding by
cooperative observation stations is a major goal of the project.
The motion understanding can be categorized into the following
three levels depending on the type of the information being processed:
- Physical Motion Analysis: The objective of this analysis is
to measure the physical object motion in the scene: a temporal
sequence of object positions, poses, and shapes are computed.
- Object Behavior Recognition: Here we mean a behavior by a class
(i.e. common pattern) of object physical motions constrained by innate
properties of an object and/or by its surrounding environments: motion
patterns such as incoming and outgoing from a door, walking, running,
bowing, and so on are recognized.
- Action Understanding: As discussed in Section 3.3,
the internal state of an embodied AVA is reflected onto the world
state, which then can be observed by other AVAs. This means that AVAs
can reason about internal states of other embodied AVAs through
observation (see equation (10)). We call
such reasoning action understanding: why an object has done and/or is
doing such an action, an intended body motion, is reasoned about.
Instead of directly attacking the most difficult action understanding
problem, we take a step by step approach from the physical motion
analysis. Moreover, since the motion understanding in the real world
includes a wide spectrum of different situations, we set several
sub-goals to clarify technical problems involved in each situation. In
what follows, we describe the sub-goals being studied in the project
with preliminary research results.
