基于误差椭圆重采样的粒子滤波跟踪算法  

Particle filter tracking algorithm based on error ellipse resampling

作　　者：徐诚[1,2] 王鑫鑫 段世红 万家旺[1,2] Xu Cheng;Wang Xinxin;Duan Shihong;Wan Jiawang(Schoo of Computer and Communication Enginering,University of Science and Technology Beijing,Bejing 100083,China;Shunde Graduate School,University of Science and Technology Bejing,Foshan 528399,China)

机构地区：[1]北京科技大学计算机与通信工程学院,北京100083 [2]北京科技大学顺德研究生院,佛山528399

出　　处：《仪器仪表学报》2020年第12期76-84,共9页Chinese Journal of Scientific Instrument

基　　金：博士后创新人才支持计划(BX20190033);广东省基础与应用基础研究基金(2019A1515110325);中国博士后基金面上项目(2020M670135);北京科技大学顺德研究生院博士后科研经费(2020BH001);中央高校基本科研业务费(06500127,FRF-GF-19-018B)资助。

摘　　要：实时可靠的导航定位是无线传感器网络的关键技术。在实际的应用环境中,定位系统误差模型一般是非线性非高斯的,传统的卡尔曼滤波算法无法提供长时间高精度的定位服务。现有研究中,粒子滤波可以处理复杂的系统模型和测量模型,但在实际应用中往往面临粒子退化和贫化问题。针对这一问题,提出了一种基于误差椭圆重采样的粒子滤波算法,用于无线传感器网络的目标定位跟踪问题。为提高粒子滤波算法在状态估计中的有效精度,在重采样过程中根据粒子的误差协方差矩阵建立不同置信水平的误差椭圆,按照粒子的几何位置进行分层,进而对不同层级的粒子进行筛选与优化,并对比计算后验克拉美罗下界性能验证该方法在累积误差优化中的有效性。实验结果表明,误差椭圆重采样粒子滤波算法精度达到1.05 m,有效改善了粒子退化和贫化问题。Real-time and reliable navigation and localization is the key technology of wireless sensor networks. In practical application environment, the error model of the positioning system is generally nonlinear and non-Gaussian, and the traditional Kalman filtering algorithm cannot provide long time and high precision positioning service. In present research, particle filter can deal with complex system and measurement models, but it often faces the problems of particle degradation and impoverishment in practical applications. Aiming at this problem, a particle filter algorithm based on error ellipse resampling is proposed, which can be used for target localization tracking in wireless sensor networks. In order to improve the effective precision of particle filter algorithm in state estimation, establish the error ellipses with different confidence levels in the resampling process according to the particle error covariance matrix, stratifies the particles according to their geometric positions, then conducts screening and optimization of the particles with different stratification levels, and verifies the effectiveness of the proposed method in cumulative error optimization through comparing with that of computing the posterior Cramer-Rao lovoer bound(PCRLB). The experiment results show that the error elliptic resampling particle filter algorithm reaches the accuracy of 1.05 m, which can effectively improve the particle degradation and impoverishment issues.

关 键 词：粒子滤波器 误差椭圆 重采样 后验克拉美罗下界 累积误差优化 

分 类 号：TH824[机械工程—仪器科学与技术]