An improved particle filter based on genetic resampling

Abstract

Abstract. The resampling of particle filter algorithm causes sample impoverishment and results in the loss of diversity. An improved particle filter algorithm is proposed by combining genetic algorithm and PF. First of all, particles are selected according to the probability with the principle of selection in genetic algorithm which make particles with larger weights to be selected more likely , and the proposed algorithm can guide the entire process to the direction of evolution. Crossover and mutation are adopted to replace the strategy of resampling which simply copy particles with high weight, delete particles with low weight and implement the particle update and optimization. Finally, the simulation experiments show that the proposed algorithm can more effectively improve the diversity of particles.

An Efficient Target Tracking Algorithm Based on Particle Filter and Genetic Algorithm

“…Therefore, the task of object tracking is highly difficult compared to the mathematical model-based tracking. To sum up, the proposed RPFGA method with fewer particles outperforms the GRPF [11] method in terms of both tracking precision and computational complexity in spite of the challenges mentioned above.…”

“…*Corresponding Author Email: smy.moghadasi@gmail.com (S. Sadegh Moghadasi) Particle filtering (PF) is the other significant technique for tracking moving objects [3,5], which is originated from Monte Carlo integration method and is mainly used in Bayesian probability estimation in nonlinear stochastic systems [5,7,11]. In the PF-based visual tracking systems, as there are only a few particles to display the probability density function (PDF) of the object's correct state, the degeneracy of particles occurs in practice [3][4][5][6].…”

“…Zhao et al [11] have presented an improved particle filter based on Genetic Resampling (GRPF) In this algorithm, crossover and mutation operators are adopted to replace the strategy of resampling. The proposed algorithm improves the particles diversity, solves the Sample Impoverishment problem and demonstrates a better performance compared to the particle filter.…”

“…In this paper, we propose a novel hybrid approach called Reduced Particle Filter Genetic Algorithm (RPFGA). Both RPFGA and GRPF [11] methods rely on genetic resampling where crossover and mutation operators perform the resampling process. It is worth mentioning that GRPF is presented with 1000 particles, while our proposed RPFGA method is presented with 20 particles.…”

“…It is worth mentioning that GRPF is presented with 1000 particles, while our proposed RPFGA method is presented with 20 particles. Moreover, performance of the GRPF hybrid method [11] has been tested on mathematical nonlinear models, while our proposed RPFGA hybrid method is applied to image sequences. Mathematical non-linear models do not encounter challenges of a real-world application as exist in visual tracking methods.…”