Adaptive block-fusion multiple feature tracking in a particle filter framework

Abstract

In this paper, we propose an adaptive block-fusion multiple feature tracking algorithm in a Particle Filter framework. The features of the object are linear weighted in a single particle filter framework by weighting their contributions using the divisibility between object and background region. A modified strategy named block-fusion is also devised to estimate the divisibility of each particle. We demonstrate the algorithm using color and gradient histogram. Finally, several experiments are implemented to verify the proposed algorithm.