I. Introduction
Many real-world applications require robust and accurate object tracking, including video surveillance, human-computer interfaces, traffic management, robot localization, etc. [1], [2]. Visual tracking can be formulated as a density estimation method (e.g., estimation of the posterior density function) [3]. Sequential Importance Sampling (SIS) is widely used as a density estimation method [3]. The associated particle filters, also known as Sequential Monte Carlo filters, have become very popular in video tracking. In particle filtering, random samples, called particles, are generated from a proposal density function and used to evaluate the importance weights, which are normalized and subsequently used to the estimate the posterior density function.