Abstract: An adaptive bandwidth object tracking method based on particle filter was proposed. Classic Particle Filter based tracking algorithm uses fixed kernel-bandwidth, as the scale changes obviously, the target may not be tracked effectively. So the principal component analysis method was introduced into the particle filtering framework to analysis the covariance matrix of the pixels within the target region. Then the most ideal tracking window including target direction and scale can be calculated. The experimental results show that the method can be adaptive to the variation of local structure of the target, moreover, spatial location and scale are good.