Kernel-based visual tracking with continuous adaptive distribution

Abstract

The template updating problem of Kernel-based tracking KBT includes two aspects: target-scale update and target-model update. The proposed algorithm can up- date both tracking window’s scale and target model by mak- ing use of continuous adaptive distribution. The ability of KBT can be complemented within its own framework with modest computation cost. The proposed tracking algorithm tries to get a balance between the stability of KBT and adaptability of CAMSHIFT for creating a robust tracker.