Student-t mixture filter for robust, real-time visual tracking

Abstract

Filtering is a key problem in modern information theory; from a series of noisy measurement, one would like to estimate the state of some system. A number of solutions exist in the literature, such as the Kalman filter or the various particle and hybrid filters, but each has its drawbacks. In this paper, a filter is introduced based on a mixture of Student-t modes for all distributions, eliminating the need for arbitrary decisions when treating outliers and providing robust real-time operation in a true Bayesian manner. © 2008 Springer Berlin Heidelberg.