Decomposition of human motion into dynamics-based primitives with application to drawing tasks. (Q1421442)

The authors assume that any human activity can be decomposed into building blocks which belong to an alphabet of elementary actions referred to as ``movemes''. They propose a formal definition for these movemes and for the classification and segmentation problem, which can be appropriately formalized in a dynamical systems framework. The core of the paper is related to the problem of segmenting data streams originating from different unknown or partially known processes which alternate in time. They propose a solution in the special case when each one of the segments is generated from a perturbed version of a linear dynamical system extracted from a finite known set of possible linear models.