Target tracking algorithm based on Mean Shift and histogram ratio background weighted

Abstract

To resolve the problem that the background pixels in an object model induce localization errors in target tracking, a new target model establishing method based on HRBW is put forward. The fuzzy membership degree based on target/background histogram log-likelihood ratio was introduced in the kernel histogram for reducing the localization errors in target tracking produced by background pixels. The method transforms only the target model but not the target candidate model and decreases the probability of target model features that are prominent in the background. The results in experiments prove that the proposed algorithm not only accelerated the convergence, but also enhanced anti-interference ability.