A fast and effective appearance model-based particle filtering object tracking algorithm

The Gaussian Mixture Model (GMM) is one of the common object representation models in the field of target tracking. However, existing GMM modeling methods are time-consuming. In this paper, we present a method to quickly model the object by a GMM in the joint feature-spatial space. A new measure based on approximations of symmetric KL-Divergence is used to compute the similarity between two GMMs. Experiments show that our modeling method is more efficient than existing methods, and our measure is more discriminative and robust than exist measures. Moreover, our tracker has better stability and a higher accuracy than the color histogram based tracker.