Edward ALTMAN
Recognition of Continuous Gestures Using Nonlinear Dynamics
Abstract:Early gesture recognition systems used discrete gestures for pointing and grasping tasks. Later systems emphasized the use of continuous gestures and sign language gestures which contain more abstract information. The thesis of this paper is that the temporal qualities of continuous gestures can be modeled more naturally using dynamical systems. The primary problem is the construction of model dynamics for gestures. A secondary problem which arises from continuous gestures is the temporal alignment, segmentation. and classification of the hand motions. The goal of this research is to exploit the multifunctionality of nonlinear systems to develop a concise mechanism for gesture recognition.
This paper describes a nonlinear technique for gesture recognition based on selective synchronization in a population of dynamical systems. A genetic algorithm is developed to learn the synchronizing dynamics. The emergent property of spatial pattern formation in a population of dynamical systems is
used to perform classification. The advantages and disadvantages of nonlinear dynamics for gesture recognition are discussed and future directions for this research are suggested at the end of the paper.