Multi-resolution tracking of a non-rigid target with particle filters for low and variable frame-rate videos

For the last decades, there has been a focus on using particle filters methods in tracking systems for their ability to solve nonlinear, non-Gaussian tracking problems. This article presents a novel tracking method with multi-resolution particle filters using color appearance model for tracking of a non-rigid target such as people. Compared to the baseline tracking by particle filters, our approach has the advantage of a two passes tracking: fast locating target state in the 1st pass and precisely refining it in the 2nd pass. After being evaluated in different kinds of people tracking environments, our system proves to offer a robust, efficient tracking performance and less computational cost.