一种改进的均方根容积粒子滤波算法

A Novel PF Approach with Squared Cubature Particles

传统的粒子滤波算法在重要性采样估计时忽略了当前量测影响。在非线性场景下,传统的粒子滤波导致个别粒子具有大权值,造成估计结果精度差。针对该问题,结合均方根容积卡尔曼滤波(SCKF)算法和Gating技术,提出了一种新的重要性函数估计算法。本算法将后验概率作为重要性采样函数,通过利用SCKF和统计距离,建立粒子与量测的关联关系,实现对重要性采样函数的均值和协方差矩阵的估计。而后,使用粒子滤波算法,对多目标状态和数目进行估计。实验表明,在非线性跟踪场景下,本算法估计精度高,估计结果稳定。

In multi-target tracking, the influence of current observations is commonly overlooked in the Importance Sampling （IS） function design of conventional Particle Filter （PF）. When targets travels with nonlinear dynamics, few particles have large weights,leading to bad estimations. To avoid such a problem,in this paper,the Squared Cubature Kalman Filter （SCKF） and a Gating method has been integrated in IS function design. To be specific, first, the posterior intensity has been selected as the IS function. Then,utilizing the SCKF and Gating method,the associations between observations and particle can be constructed to estimate the mean and covariance matrix. On this basis,by means of the particle filter,states and numbers of multi-target can be estimated. It has been proved that the proposed approach has more accuracy and stability than the convention PF in nonlinear multi-target scenario.