EVENT RECOGNITION WITH TIME VARYING HIDDEN MARKOV MODEL

Standard Hidden Markov Model (HMM) and the more gen-eral Dynamic Bayesian Network (DBN) models assume sta-tionarity of state transition distribution. However, this as-sumption does not hold for many real life events of interest. In this paper, we propose a new time sequence model that extends HMM to time varying scenario. The time varying property is realized in our model by explicitly allowing the change of state transition density as the time spent in a partic-ular state passes by. Rather than keeping transition densities at different time spots independent of each other, we exploit their temporal correlation by applying a hierarchical Dirichlet prior. This leads to a more robust time varying model, espe-cially when training data are scarce. We also employ Markov Chain Monte Carlo (MCMC) sampling in learning the MAP estimate of time varying parameters, with a transition kernel incorporating linear optimization. The proposed model is ap-plied to recognizing real video events, and is shown to out-perform existing HMM-based methods. Index Terms — event recognition, time varying, HMM, MCM