|
Major Research Subjects of Ikeuchi, K. (1998-2000) Professor
Further information is available at http://www.cvl.iis.u-tokyo.ac.jp/~ki/
- Automatic generation of mixed reality systems
Application of mixed reality systems to a wide range of field
applications is presently anticipated. However, programmers manually
create most of the models for these systems. The virtual objects
and virtual space in applications such as virtual shopping and
virtual museum guides are based on an actual object and space.
If a system could be built to observe these actual objects/space
and automatically create models for mixed reality systems, manual
programming labor as well as modeling costs could be drastically
decreased. To create such an automatic system, techniques such
as methods to obtain the geometric figuration (e.g. dimension,
curvature) in actual space (environment) and capture texture such
as reflectivity and color are currently subject to research and
development.
- Intelligent Transportation System (ITS)
Until today, automatic driving robots have been designed as an
autonomous robot independent of humans. With the arrival of the
21st century, robots (intelligent cars) and humans, as well as
the robots(intelligent cars) surrounding them, will need to be
able to intelligently move about in coordination with the Intelligent
Transportation System of the new century. Development of a robot(intelligent
car) capable of understanding the movements of humans and other
robots surrounding them, comparing the surrounding road environment
as well as information from a higher control system is under research.
Issues to be researched include: understanding human motion from
a sequence of input images; decision making by comparing map information
and the surrounding situation; adding information and map information
onto live images.
- Automatic generation of object recognition programs
Programmers have been manually coding object recognition programs
for a computer to identify objects. In object recognition application
scenarios, there is a CAD model of these objects. We have been
developing techniques to automatically convert object models into
recognition programs. Research issues include how to extract useful
features from object models, how to determine these features to
recognize one object from another and how to generate recognition
algorithms from these features.
- Robotics learning from observation
Taking a child as an example, human behavioral learning occurs
through visual observation. If a computer had the same capability,
it could acquire the ability to perform the necessary tasks without
being manually programmed through observing a human demonstrate
these tasks in front of it. Our research aims to construct such
a robot. Current research topics include: how to divide a sequence
of input images of human movement into meaningful segments; analyze
each segment and extract movement primitives; mapping primitives
onto robot actions.
|
