A fast object tracking approach based on the motion vector in a compressed domain

Abstract

Particle set sampling and weighting are both at the core of particle filter-based object tracking methods. Aiming to optimally represent the object-s motion state, a large amount of particles - in the classical particle method - is a prerequisite. The high-cost calculation of these particles significantly slows down the convergence of the algorithm. To this problem, a prior approach which originated from the process of video compressing and uncompressing is introduced to optimize the phase of particle sampling, making the collected particles centre on and cover the object region in the current image. This advantage dramatically reduces the number of particles required by the regularized particle sampling method, solving the problem of the high computational cost for tracking objects, while the performance of the algorithm is stable. © 2013 Wang et al.