Decomposition of Human Motion into Dynamics Based Primitives with Application to Drawing Tasks
D. Del Vecchio, R. M. Murray and P. Perona
Using tools from dynamical systems and systems identiﬁcation we develop a framework for the study of primitives for human motion, which we refer to as movemes. The objective is understanding human motion by decomposing it into a sequence of elementary building blocks that belong to a known alphabet of dynamical systems. In this work we address the problem of deﬁning conditions under which collections of signals are well-posed according to a dynamical model class M and then can generate movemes. Based on the assumption of well-posedness, we develop segmentation and classiﬁcation algorithms in order to reduce a complex activity into the sequence of movemes that have generated it. Using examples we show that the deﬁnition of well-posedness can be applied in practice and show analytically that the proposed algorithms are robust with respect to noise and model uncertainty. We test our ideas on data sampled from ﬁve human subjects who were drawing ﬁgures using a computer mouse. Our experiments show that we are able to distinguish between movemes and recognize them even when they take place in activities containing more than one moveme at a time.
- Preprint: http://www.cds.caltech.edu/~murray/preprints/dmp03-automatica.pdf
- Project(s): Template:HTDB funding::NSF