Visual tracking with singular value particle filter

Robust tracking is an important and challenging problem in computer vision. Most existing algorithms do not work well if there are confusing objects in the surrounding environment or the target appearance has a significant change. This paper describes a novel particle filter for object tracking. First, we treat the blob image of the object as a matrix and adopt singular values to construct the feature model. In the second stage, the particle filter scheme is applied for tracking. According to particle degeneracy Metropolis-Hastings sampling is proposed to obtain more efficient particle filter. Borne out by experiments, we demonstrate the proposed method is able to track well under scale variation and when there are confusing objects in the background. Besides, it has higher performance than conventional particle filters in terms of weight and number of particle.