计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (22): 84-91.DOI: 10.3778/j.issn.1002-8331.2301-0086

• 理论与研发 • 上一篇    下一篇

高斯过程回归泊松多伯努利衍生滤波器

宋营营,宋骊平   

  1. 西安电子科技大学 电子工程学院,西安 710071
  • 出版日期:2023-11-15 发布日期:2023-11-15

Gaussian Process Regression Poisson Multi-Bernoulli Filter with Target Spawning

SONG Yingying, SONG Liping   

  1. School of Electronic Engineering, Xidian University, Xi’an 710071, China
  • Online:2023-11-15 Published:2023-11-15

摘要: 针对伽马高斯逆威舍特混合泊松多伯努利(Gamma Gaussian inverse Wishart mixed Poisson multi-Bernoulli,GGIW-PMB)滤波器无法估计非椭圆形状目标的问题,提出了将泊松多伯努利滤波器与高斯过程回归模型结合的方法,可对非椭圆形状目标进行准确估计。考虑到衍生存在情形下无法有效提取衍生目标及其扩展形状的问题,提出了一种衍生目标检测及建模方法,通过量测数的变化来对衍生事件做出假设,根据真实场景关系计算衍生目标状态,实现衍生目标的检测和跟踪。在泊松多伯努利滤波器的基础上,采用高斯过程回归模型作为量测模型,结合所提衍生模型,提出了基于高斯过程回归的泊松多伯努利衍生(Gaussian process regression Poisson multi-Bernoulli filter with target spawning,GPR-PMBS)滤波器。仿真结果表明,GPR-PMBS滤波器相比于GGIW-PMB滤波器能更为准确地估计非椭圆形状目标,并且在衍生存在的情形下,也可以有效提取出衍生目标及其形状,在有衍生情况的扩展目标跟踪场景中表现出良好性能。

关键词: 高斯过程回归, 泊松多伯努利, 衍生目标, 扩展目标跟踪

Abstract: In order to solve the problem that the Gamma Gaussian inverse Wishart mixed Poisson multi-Bernoulli (GGIW-PMB) filter cannot estimate the non-elliptic shape target, a method combining the Poisson multi-Bernoulli filter and Gaussian process regression model is proposed to estimate the non-elliptic shape target accurately. In view of the problem that the spawned sub-extended target and its extended shape cannot be extracted effectively, a spawned sub-extended target detection and modeling method is proposed. The spawned sub-extended target is detected by the change of the measurement number, and its state is calculated according to the real scene relationship. Combined with the proposed spawned sub-extended model, Gaussian process regression based Poisson multi-Bernoulli filter with target spawning(GPR-PMBS) is proposed. The simulation results show that compared with GGIW-PMB filter, the proposed algorithm can estimate the non-elliptic shape target more accurately, and can extract the spawned sub-extended target and its shape effectively, which shows good performance in the extended target tracking scenarios with target spawning.

Key words: Gaussian process regression, Poisson multi-Bernoulli, target spawning, extended target tracking