Facial Features Tracking Using Auxiliary Particle Filtering and Observation Model Based on Bhattacharyya Distance

Abstract

Recently, particle filtering has become an effective algorithm for facial feature tracking. One problem with particle filtering is that as the dimensionality of the state space increases, a large number of particles that are propagated from the previous time are wasted in area where they have low observation probability, hence a very large number of particles are necessary to track the state and, as a result the complexity increases and the speed of algorithm reduces. In this research, auxiliary particle filter with factorized likelihoods is used in order to overcome this problem. In a tracking approach, the estimated state is updated by incorporating the new observations. Therefore an observation model is needed. In this paper a novel color-based observation model that is invariant to changes in illumination intensity is proposed. The proposed observation model employs the Bhattacharyya distance to update a prior distribution calculated by the particle filter. In this paper experimentally is showed that the proposed algorithm clearly outperforms multiple independent template tracking.