面向非线性MTT的多模型泊松多伯努利混合滤波算法

Algorithm of Multi-model Poisson Multi-bernoulli Mixture Filter for Nonlinear MTT

在多目标跟踪(Multi-target Tracking,MTT)的非线性特性与低检测概率情况下,针对多伯努利滤波算法的高斯混合(Gaussian Mixture,GM)实现难以精确估计目标的势与运动状态的实际问题,本文提出了一种适用于非线性系统的泊松多伯努利混合滤波(Poisson Multi-Bernoulli Mixture Filter,PMBM)算法.首先,推导出多模型泊松多伯努利混合滤波的高斯混合(GM Multi-model PMBM,GM-MM-PMBM)实现过程.然后,分别对GM-MM-PMBM的伯努利高斯分量进行预测与更新,实现了基于非线性系统的MTT.为提升系统稳定性,基于平方根协方差矩阵推导出GM-MM-PMBM均方根容积卡尔曼滤波算法的实现过程.最后,仿真实验综合验证了本文算法的跟踪性能.

Aiming at the problem that the multi-Bernoulli filter implemented by Gaussian Mixture(GM)is difficult to accurately estimate the cardinality and states of targes under the nonlinear model and low detection probability in multi-target tracking(MTT),a realization method of Poisson multi-Bernoulli mixed filter under nonlinear system model was proposed.First the recursive process of GM realization of Multi-model Poisson multi-Bernoulli mixed filter(MM-PMBM)is derived.Afterwards the Gaussian components during the transferring process are used to predict and update in order to achieve MTT in nonlinear systems.For enhancing numerical stability,the square root covariance matrix in the recursive process is directly transferred,and then the GM-MM-PMBM filter is explored.Further the realization of the square-rooted cubature Kalman filter is achieved.Finally the tracking performance of proposed algorithm is comprehensively verified by simulation experiments.