Particle Filter with Hybrid Proposal Distribution for Nonlinear State Estimation

Abstract

Particle filters have been widely used in solving nonlinear filtering problems. Proposal Distribution design is a key issue for these methods and has vital effect on simulation results.  Various proposal distributions have been proposed to improve the performance of particle filters, but practical situations have promoted the researchers to design better candidate for proposal distributions in order to gain better performance. This paper proposes a hybrid proposal distribution designed for particle filtering framework. The algorithm uses hybrid Kalman filter to generate the proposal distribution, which make efficient use of the latest observations and generate more closed approximation of the posterior probability density. The yielded algorithm is named as hybrid Kalman particle filter. In the experiments, we use a scalar estimation model and a real world application problem to evaluate the new algorithm. The experimental results show that the new algorithm outperforms other algorithms with different proposal distributions. In order to reduce the time consumption of the new algorithm, we propose an improvement strategy, namely partition-conquer strategy, in which the particles are partitioned into two parts, with one part drawn from the hybrid Kalman filter, another part from the transition prior. The validity of the strategy is verified through an experiment

Double hybrid Kalman filtering for state estimation of dynamical systems

“…Below, there are presented Doubled Modified Hybrid Kalman Filter (dHKFmod) and Doubled Modified Hybrid Kalman Particle Filter (dHKPFmod) algorithms. They are based on HKFmod and HKPFmod from [1], where for double methods based on traditional HKF and HKPF [2], the order of EKF and UKF algorithms inside were reversed. k|k) , P EKF1 (k|k) , y (k) ]…”

“…Very important branch of science, especially in the noisy environment of measurements, is state estimation. This paper develops the problem of state estimation of dynamical systems and refers to research from [1], where authors proposed modification of Hybrid Kalman Filter and Hybrid Kalman Particle Filter [2].…”

Particle Filter and Visual Tracking: a Hybrid Resampling approach to improving robustness in cluttered and occluded environments

“…But it cannot be used efficiently in situations when occlusions and cluttered environments happen, due to the fact that we cannot evaluate the probability density function (PDF) before the particles are drawn. Many works (as [23], [24] and [25]) have focused their efforts on suboptimal approaches, because the transition prior based particle filter (generic particle filter) is easy to implement and has been widely used to solve problems in real-world scenarios, especially in visual tracking. Several versions of the generic particle filter, from the Condensation algorithm to the Unscented transformation, failed in complex tracking scenarios due to the sample impoverishment problems.…”

An Improved Unscented Particle Filter with Global Sampling Strategy

“…The proposed algorithm is also compared with the IUPF method presented in [11]. In IUPF, the proportion of the particles generated from UKF is = 20%, 50%, and 80%, and the other particles are generated from the transition prior, respectively.…”