Prototype-based Representation of Complex Human Movements: Studies in Brains and Machines
Speaker: Martin Giese , Dept. of Cognitive Neurology, University Clinic Tübingen
Date: March 24 2004
Time: 4:00PM
Location: E25-401
Contact: Mary Pat Fitzgerald, 3-0551, marypat@ai.mit.edu
Relevant URL:
The efficient representation of complex human movements is an important problem for visual perception and many technical applications. Inspired by previous work on the view-based encoding of complex shapes, the talk will present a number of results that explore representations of complex human movements that are based on learned prototypical examples.
First, a learning algorithm is presented that represents complex human action sequences by linear combination of learned prototypical example trajectories. The method decomposes action sequences automatically into movement primitives and works with very small amounts of training data. It provides an intuitive parameterization of classes of complex human movements, and has interesting applications for movement synthesis (e.g. in psychophysics), as well as for movement analysis (e.g. for the automatic estimation of skill levels in sports, or the quantitatification of movement disorders of neurological patients).
Second, evidence is presented that supports the idea that human visual recognition of biological motion might be based on learned prototypes. A neural model is proposed that accounts for a variety of known phenomena in biological motion perception, using neural mechanisms that are consistent with well-established neuroanatomical and neurophysiological facts. The model motivates learning experiments with biological motion stimuli. Key results from such experiments and a related fMRI study will be presented, providing some insight in possible neural correlates of the learning process.
See other events that are part of Brains & Machines Seminar Series Spring 2004
See other events happening in March 2004
