Adaptive Object Tracking Based on an Effective Appearance Filter
Colour histogram based particle filter is an efficient technique for tracking. The scale of kernel is a crucial parameter which is formed by the weighted particles. However, changing the scale simply by the histogram similarity sometimes results in the kernel fast shrinks into local optimal. Therefore, a method is proposed to detect the potential hints of shrinking first. Then, the centre distribution of particles is utilized to resize the kernel. The similarity likelihood between modified kernel and the ground truth is used for forming the resampling step to solve the degeneracy problem. We test the proposed approach on both simulated and real scenarios. In these experiments, the method can efficiently solve the problem of shrinking.