Iterative particle filter for visual tracking

• The paper presents a novel iterative PF (IPF).
• The search scope is iteratively contracted.
• The contracted search scope can sufficiently utilize particles.
• The contracted search scope can eliminate the influence of the clutter.
• IPF can achieve higher tracking accuracy at the same computational burden.

Particle filter (PF) has been the subject of considerable attention in visual tracking. How to approach the true target state with computation cost as low as possible has always been an important issue. In this paper, a novel iterative PF (IPF) is proposed, which can converge to the true target state as close as possible by sampling the particles iteratively with the search scope contracted. The search scope is iteratively contracted around the centers determined by the previous converging results. Compared with annealed PF (APF) and PF, IPF can converge much closer to the true target state, thus improvement in sampling efficiency, clutter elimination, and tracking accuracy with the same computational burden.