基于GLMB滤波和Gibbs采样的多扩展目标有限混合建模与跟踪算法

本文针对杂波条件下多扩展目标的状态估计, 目标个数估计, 扩展目标形状估计问题, 提出了一种基于标签随机有限集(Labelled random finite sets, L-RFS)框架下多扩展目标跟踪学习算法, 该学习算法主要包括两方面:多扩展目标动态建模和多扩展目标的跟踪估计.首先, 结合广义标签多伯努利滤波器(Generalized labelled multi-Bernoulli, GLMB)建立了扩展目标的量测有限混合模型(Finite mixture models, FMM), 利用Gibbs采样和贝叶斯信息准则(Bayesian information criterion, BIC)准则推导出有限混合模型的参数来对多扩展目标形状进行学习, 然后采用等效量测方法来替代扩展目标产生的量测, 对扩展目标形状采用椭圆逼近建模, 实现扩展目标形状与状态的估计.仿真实验表明本文所给的方法能够有效跟踪多扩展目标, 并且在目标个数估计方面优于CBMeMBer算法.此外, 与标签多伯努利滤波(LMB)计算比较表明: GLMB和LMB算法滤波估计精度接近, 二者精度高于CBMeMBer算法.

A Modeling and Tracking Algorithm of Finite Mixture Models for Multiple Extended Target Based on the GLMB Filter and Gibbs Sampler

In this paper, a new multiple extended target tracking learning algorithm based on labelled random finite sets (L-RFS) framework is proposed to estimate the number, shape and state of extended targets under clutter conditions. The algorithm mainly includes two aspects: multi-extended target dynamic modeling and multi-extended target tracking estimates. Firstly, a finite mixture model (FMM) of extended target is established under the generalized labelled multi-Bernoulli (GLMB) filter. Learning the parameters of finite mixture model by Gibbs sampling and Bayesian information criterion (BIC), and then equivalent point target measurements are used in place of the actual extended target measurements. Finally, the proposed ellipse approximation model is used to realize the estimation of the extended target shape. The simulation results show that the proposed algorithm can effectively track the multiple extended targets and it is superior to CBMeMBer algorithm in the estimation of the number of extended targets. In addition, comparison with LMB filter shows that: The filtering accuracy of the GLMB and LMB algorithms are close to each other, and the accuracy of both algorithms is higher than CBMeMBer algorithm.