|
|
Goal Achievement Based on Action Modeling and Recognition for Virtual and Real Agents |
The focus of this project is the concept of action which can basically
be seen as a unit motion. Synthetic agents are animated by the application
of actions and the recognition of the live performers' actions. We are
especially interested in the action modeling and recognition level in
order to build real-time interactive and cooperative environments.
This project is mainly based on human posture and gesture recognition
to be used in interfacesbetween real and virtual environments. We use a
magnetic motion capture
system, Ascension Flock of Birds and an anatomical converter (T.Molet).
Our work is based on the AGENTLIB library, born in the context of the European project ESPRIT Humanoid II. It allows us to easily apply motion generators to synthetic agents, either humanoids or other. Several motion generators can be applied on mutually exclusive bodyparts. An automatic transition between motions is performed on the intersection of overlapping sets of animated bodyparts.
The humanoids are autonomous and decide themselves where and how to
grasp the object. A special collision detection algorithm avoids finger-object
interpenetration. Current improvements are bending of the body and dynamic
considerations to achieve the grasping.
Emering L., Boulic R., Balcisoy S., Thalmann D., Real-Time Interactions with Virtual Agents Driven by Human Action Identification, First ACM Conf. on Autonomous Agents'97, Marina Del Rey, 1997
Emering L., Boulic R., Balcisoy S., Thalmann D., Multi-Level
Modelling and Recognition of Human Actions Involving Full Body Motion,
First ACM Conf. on Autonomous Agents'97, Marina Del Rey, 1997
Contact: Luc Emering
