Particle Filter Algorithm Based on Adaptive Resampling Strategy
Zhaoying Wang,Zhentao Liu,Weiqun Liu,Yunbo Kong
DOI: https://doi.org/10.1109/emeit.2011.6023752
2011-01-01
Abstract:To solve the problem of particle degradation in particle filter algorithm, an adaptive tracking algorithm based on particle grouped optimization is proposed. Particle filter has been greatly developed in tracking field because of its ability of maintaining state distribution with multi-modal and robustness to noise. However, the conventional particle filter has some deficiencies, such as high computational cost and low sampling efficiency. In addition, the complexity of tracking scenes poses great challenge on tracking algorithm. On the basis of conventional particle filter algorithm, the paper is improved by the establishment of feature histogram and particle resampling strategy. Considering the conspicuousness and similarity of target and background, a ratio relation is set up to select the feature which can differentiate the prospect target and background to its extent, and the number of interval of selected feature is determined by weighted discrimination. By analyzing particle space distribution, a novel resampling strategy is proposed to adjust the number of particles and particle relative positions adaptively by duplication, linear combination and elimination, which optimizes particle performance. The effectiveness of the proposed algorithm is demonstrated by simulation.