两级信息融合协同采样粒子滤波及应用

项目名称： 两级信息融合协同采样粒子滤波及应用

项目编号： No.61473227

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 自动化技术、计算机技术

项目作者： 左军毅

作者单位： 西北工业大学

项目金额： 80万元

中文摘要： 粒子滤波因其估计结果具有渐进最优性而成为非线性滤波领域的热点方法。然而面对复杂系统，其性能可能会因粒子在状态空间分布欠合理而恶化。本项目以目标跟踪和飞行器姿态估计为背景，从重采样和重要性采样两个方面研究粒子滤波的高效采样方法体系，通过重要性采样和量测更新两个层次上的信息融合来实现鲁棒高精度状态估计。主要内容包括：在重采样方面，研究能实现粒子贫化和粒子退化有效折中的力度可控重采样技术；在重要性采样方面，研究将粒子配置在高似然区域的快速采样方法，并在此基础上设计多模自适应重要性采样策略，实现先验信息和量测信息导引下的协同采样；分别在集中式和分布式融合结构下探索多传感器协同鲁棒采样理论，研究能够在采样阶段提前利用多传感器信息的两级信息融合理论及算法。项目的实施能丰富粒子滤波的理论基础，并对其走向工程应用起到推动作用。

中文关键词： 状态估计；信息融合；目标跟踪；粒子滤波；重要性采样

英文摘要： Particle filter has become an active research topic in the field of nonlinear filtering due to the asymptotic optimality of its estimate. However, in dealing with the complex dynamic system, its performance may deteriorate because of the unreasonable distribution of particles in state space. This project, set against backdrops of target tracking and attitude estimation of flight vehicles, aims to develop the high-efficiency sampling method for particle filter in two aspects, i.e. resampling and importance sampling, and also to achieve robust high-precise state estimate by using the strategy of multi-source information fusion both in the stage of importance sampling and in the stage of measure update. The main contents include: by establishing the concept of strength controllable resampling, a new resampling technique will be studied with the aim of reaching the desirable tradeoff between the particle degradation and the particle impoverishment; a high-efficient sampling method will also be studied for the porpuse of placing the particles on the high likelihood region of a specific sensor. Based on the sampling method, an adaptive multi-mode importance sampling strategy will be designed to realize the collaborative sampling according to the prior information and measurement information; The multi-sensor robust sampling theory will be investigated in the centralized fusion structure and distributed fusion structure respectively, and, also, the two-stage information fusion theory and algorithm will be studied, which has the ability to use the multi-sensor information in the stage of importance sampling in advance.The implementation of the project can not only enrich the theory of particle filter, but also promote its engineering application.

英文关键词： state estimate;information fusion;target tracking;particle filter;importance sampling