抄録
This paper proposes the mimesis from 2D partial observation of vector field. Its framework is HMM based mimesis model for human motion understanding. The algorithm is extended into motion understanding from monocular image sequences (eg., unlabeled markers information or marker-less texture information). Various conversions (by kinematics, coordinate transformation, and projection) of symbols are processed. With the particle filter of the transformation matrix, the humanoid estimates human pose, which is the transformation matrix between the human's coordinate and humanoid's coordinate. Vector field is used for natural motion understanding.
