高斯过程泊松多伯努利混合滤波算法及其变分优化

Gaussian process Poisson multi-Bernoulli mixture filtering and its variational optimization

中文关键词  目标跟踪  泊松多伯努利混合滤波  高斯过程  变分贝叶斯优化

英文关键词  target tracking  Poisson multi-Bernoulli mixture filtering  Gaussian process  variable Bayesian optimization

中文摘要

      针对现有算法对多扩展目标跟踪精度低的问题, 本文提出了一种高斯过程泊松多伯努利混合(GP-PMBM)滤波算法及其变分优化. 首先, 基于高斯过程原理建立了增广状态空间模型, 接着, 将其与泊松多伯努利混合滤波器相结合, 提出GP-PMBM算法. 然后, 针对因使用非线性滤波技术而导致GP-PMBM滤波精度下降的问题, 使用变分贝叶斯优化更新结果, 实现了对目标状态的优化更新, 提升了滤波器的估计精度. 仿真结果表明, 与已有的滤波算法相比, 所提算法具有更高的跟踪精度, 并且, 在只有部分量测的场景中跟踪性能更稳定.

英文摘要

      In response to the problem of low tracking accuracy for multi-object tracking in existing algorithms, a Gaussian process Poisson multi-Bernoulli mixture (GP-PMBM) filtering algorithm and its variational optimization are proposed. Firstly, an augmented state space model is established based on the principles of Gaussian processes. Subsequently, to address the issue of decreased filtering accuracy in GP-PMBM caused by the use of nonlinear filtering techniques, variable Bayesian optimization is utilized to update the results, achieving optimized updates of the target states and enhancing the estimation accuracy of the filter. Simulation results demonstrate that the proposed algorithm has higher tracking accuracy compared to existing filtering algorithms and exhibits more stable tracking performance in scenarios with only partial measurements.