An object tracking algorithm based on multi-model and multi-measurement cues

In this paper, we present a new visual object tracking algorithm for video surveillance systems. The main contribution of this paper is the development of a new particle filter (PF) that incorporates multiple dynamic models and multiple measurement cues to achieve reliable and accurate tracking in different tracking scenarios. More specifically, this algorithm utilizes a discretized proposal distribution to obtain more support from the system posterior distribution. In addition, a new likelihood model is designed to take advantage of multiple measurement cues to achieve reliable estimation. Also, this algorithm is implemented in the multiple model framework to further improve the robustness. Experimental results have demonstrated that this new algorithm is capable to provide effective and reliable tracking results in different tests.