An improved Camshift-based particle filter algorithm for face tracking

Abstract

To improve the poor performance of face tracking due to complex illumination, cluttered background and poses variations, an improved Camshift-based particle filter algorithm is proposed in this paper. First, under the particle filter framework, we propose a novel feature extraction method called the block rotation-invariant uniform local binary pattern (BriuLBP), and combine with color features to represent the appearance model of face in tracking tasks. Compared to the traditional LBP, the BriuLBP method has the advantage of exploring space structure to enhance the accuracy of tracking, in term of preserving the robustness to illumination and appearance variations. Furthermore, in order to maintain the balance between sampling effectiveness and diversity, and to avoid Camshfit iterative to end at local optimal, a method of adjusting the number of particles in Camshift according to the target changing is proposed. Experimental results show that the proposed method performs better than the traditional particle filter algorithm based on Camshift in both robustness and accuracy, and achieve faster and more robust tracking for face tracking. © 2012 Springer-Verlag.