A novel particle filter for tracking fast target

Abstract

This paper proposes a fast target tracking method in which particle filter is improved using Gaussian kernel and evolutionary strategy. We use Gaussian kernel function to replace the Dirac kernel function, which can decrease the degeneracy problem of the traditional particle filter partly. To further improve the performance of particle filter, we introduce evolutionary strategy into the process of Gaussian kernel particle filtering. It uses only mutation operation, which has less computation than genetic algorithm. And it can prevent the impoverishment problem and steer the particles towards local mode of posterior probability effectively. The proposed method can track fast target robustly using fewer particles than the standard particle filter and Gaussian kernel particle filter.