Recognizing human behavior through nonlinear dynamics and syntactic learning

This work applies nonlinear dynamics to model the encoded time series describing human activities performed in the spatio-temporal domain. We augment the concepts of symbolic dynamics and formal language theory to pattern recognition in generating probabilistic feature extraction which are used to drive a Bayesian classifier for behavior recognition. Our motivation for using stochastic context-free grammar (SCGF) is to aggregate low-level events detected so that we can construct higher-level models of interaction. This is a novel attempt in coupling symbolic dynamics with a stochastic model to construct a spatio-temporal ordered SCFG. Extended statistical tests and comparative analysis with Bayesian classification and k-nearest neighbour (K-NN) classification of time series sequences demonstrate a superiority of the proposed method of gesture recognition using concepts from nonlinear dynamics.