conference paper
Multi-level modeling and recognition of human actions involving full body motion
1997
Proceedings of the First International Conference on Autonomous Agents
A multi-level model of human actions involving full body motion is proposed. Actions are considered as the combination of action primitives at different levels: motions of the center of mass, end effectors, and final postures of the virtual skeleton. The model is especially designed for fast action recognition from the input of the human motion at the joint level