Video background subtraction using online infinite dirichlet mixture models

Video background subtraction is an essential task in computer vision for detecting moving objects in video sequences. In this paper, we propose a novel Bayesian nonparametric statistical approach to subtract video background. The proposed approach is based on a mixture of Dirichlet processes with Dirichlet distributions, which can be considered as an infinite Dirichlet mixture model. Compared to other background subtraction approaches, the proposed one has the advantages that it is more robust and adaptive to dynamic background, and it has the ability to handel multi-modal background distributions. Moreover, thanks to the nature of nonparametric Bayesian models, the determination of the correct number of components is sidestepped by assuming that there is an infinite number of components. Our results demonstrate the merits of the proposed approach.