粒子滤波理论及其在水下目标跟踪问题中的应用
文献类型：学位论文
| 作者 | 刘斌 |
| 学位类别 | 博士 |
| 答辩日期 | 2009-01-22 |
| 授予单位 | 中国科学院声学研究所 |
| 授予地点 | 声学研究所 |
| 关键词 | 数据关联 扩展卡尔曼滤波 不敏卡尔曼滤波 粒子滤波 状态/参数估计 水下目标跟踪 序列蒙特卡罗 纯方位目标跟踪 杂波 传感器时序 协方差矩阵控制 二维分配 奇异值分解 |
| 其他题名 | Particle Filtering Theory with Applications in Underwater Target Tracking |
| 学位专业 | 信号与信息处理 |
| 中文摘要 | 状态估计（也被称作滤波、跟踪、信任更新）是指利用一系列量测数据来计算我们感兴趣目标所处的状态，它是所有智能系统的核心问题。本论文以自上世纪90年代以来发展迅速的序列蒙特卡罗方法（也称作粒子滤波）作为主要的数学工具，致力于解决具有挑战性同时也是非常重要的水下目标跟踪问题。 第一章交代了本文的研究背景、国内外相关研究现状，并提出本文的创新点。 第二章首先回顾经典的卡尔曼滤波、扩展卡尔曼滤波算法。介绍了最近发展起来的不敏卡尔曼滤波器，它是基于数值算法无迹变换的最新发展起来的一种固定采样算法。通过理论分析和计算机仿真实验对扩展卡尔曼滤波和不敏卡尔曼滤波进行了对比。具体地，针对纯方位跟踪问题，提出扩展卡尔曼和不敏卡尔曼折中算法。新方法仅利用与扩展卡尔曼滤波相近的计算量即能实现与不敏卡尔曼滤波器相当的估计性能。 第三章给出贝叶斯理论架构，并对粒子滤波器原理进行了简介。本章研究一种参数估计问题，即频率跟踪，和一维状态跟踪问题。分别设计了相应的粒子滤波算法。对于前一问题，设计了基于高斯核和Metropolis-Hastings马尔可夫链蒙特卡罗技术的粒子滤波器；对于后者，提出一种利用奇异值分解卡尔曼滤波器产生建议分布的粒子滤波器。通过计算机仿真验证了所提方法至少对于这些具体问题是优于现有方法的。 从第四章开始我们着眼于本论文的主体内容-具有挑战性的水下目标跟踪问题。具体来说，第四章针对单目标跟踪问题，提出一种新的类检测前跟踪纯方位目标跟踪架构；第五章着眼于杂波环境下的多目标跟踪，提出基于不敏粒子滤波器和二维分配算法的跟踪方法；第六章处理“智能”目标跟踪问题，在粒子滤波器框架内利用一种传感器时序管理算法-协方差矩阵控制技术来解决此问题。 最后一章总结本论文，并给出未来该领域的研究方向。 |
| 英文摘要 | State estimation (also known as filtering, tracking, belief update, and situation assessment) is the problem of figuring out the state of the object which we are interested in, given a sequence of percepts. It is a core problem for all intelligent systems. This thesis targets for solving the problem of underwater target tracking, which is challenging also very important. The involved methodology is mainly the Sequential Monte Carlo (SMC) method, also called the particle filter (PF), which has been developed fast since the 90th of last century. In the 1st chapter we describe the research background of this thesis, present current status of related work in the worldwide area, and summarize the main innovative points of this thesis. The 2nd chapter review the classical filtering methods, i.e. Kalman filter, extended Kalman filter (EKF). Then we presente a newly developed method called Unscented Kalman filter (UKF), which is a determinate sampling method based on a numerical technology called Unscented Transform (UT). We compare the EKF and UKF both in theoretical aspect and by way of computer simulations. Specifically, for a Bearings-only-tracking problem, we propose a tradeoff method between EKF and UKF, which can approximately achieve the performance of the UKF with similar computational loads as the EKF. In the 3rd chapter, we present the Bayesian theoretical framework at first then give a brief introduction for the PF. This chapter focuses on a parameter estimation problem, i.e. frequency tracking, and a one-dimensional state tracking problem, for both of which, a specific designed PF is used. For the former, we design a PF based on Gaussian Kernel and Metropolis-Hastings Markov Chain Monte Carlo (MCMC) move step; for the latter, a new PF method which uses a batch of singular value decomposition (SVD) Kalman filter to obtain the proposal distributions is proposed. Both of these methods are superior to existing methods at least for the specific problems. From the 4th chapter on, we turn to the mainbody of this thesis, i.e. challenging underwater target tracking problems. Specifically，the 4th chapter focuses on single target tracking and a new Track-Before-Detect-like Bearings-only-tracking scheme is proposed; the 5th chapter treats multi-target tracking in cluttered observation environment and a new method based unscented Particle filter (UPF) and 2D-Assignment is proposed; the 6th chapter deals with “smart” target tracking, wherein we resort to a sensor scheduling technique called covariance control and embed it in the particle filter framework to resolve the tracking of a smart target. In the last chapter, we summarize this thesis before giving a future avenue of research in this area. |
| 语种 | 中文 |
| 公开日期 | 2011-05-07 |
| 页码 | 110 |
| 源URL | [http://159.226.59.140/handle/311008/506]  |
| 专题 | 声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 刘斌. 粒子滤波理论及其在水下目标跟踪问题中的应用[D]. 声学研究所. 中国科学院声学研究所. 2009. |
入库方式： OAI收割
来源：声学研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。