Improved differential evolution-based particle filter algorithm for target tracking

This paper presents an improved target tracking algorithm based on the differential evolution particle filter (DEPF) in order to solve the problem of particle degeneracy. In this method, the mutation, crossover and selection steps of the differential evolution are introduced into the particle filter, which optimizes the particle sampling process. Particles are moved towards regions where they are close to the value of the posterior density function through DEPF algorithm, and it improves the estimation precision, inhibits the particle degeneracy and enhances the diversity of the particles. The target observation model based on the color feature is constructed simultaneously in the algorithm, and the method of the target model update in different states is also studied. The experimental results show that the target even in attitude change and occlusion can be tracked accurately and real-timely with the DEPF algorithm. The proposed algorithm has higher tracking accuracy and is robust.