Bayesian Learning of Infinite Asymmetric Gaussian Mixture Models for Background Subtraction

Abstract

Background subtraction plays an important role in many video-based applications such as video surveillance and object detection. As such, it has drawn much attention in the computer vision research community. Utilizing a Gaussian mixture model (GMM) has especially shown merit in solving this problem. However, a GMM is not ideal for modeling asymmetrical data. Another challenge we face when applying mixture models is the correct identification of the right number of mixture components to model the data at hand. Hence, in this paper, we propose a new infinite mathematical model based on asymmetric Gaussian mixture models. We also present a novel background subtraction approach based on the proposed infinite asymmetric Gaussian mixture (IAGM) model with a non-parametric learning algorithm. We test our proposed model on the challenging Change Detection dataset. Our evaluations show comparable to superior results with other methods in the literature.