水下机动目标跟踪关键技术研究
文献类型：学位论文
| 作者 | 李为
|
| 学位类别 | 博士 |
| 答辩日期 | 2015-05-29 |
| 授予单位 | 中国科学院沈阳自动化研究所 |
| 授予地点 | 中国科学院沈阳自动化研究所 |
| 导师 | 封锡盛 ; 李一平 |
| 关键词 | 混合去偏量测转换 解耦 实时辨识自适应滤波 幅值信息 集成概率数据关联算法 |
| 其他题名 | Research on key technology of maneuvering underwater target tracking |
| 学位专业 | 模式识别与智能系统 |
| 中文摘要 | 随着近几十年的发展，自主水下机器人（Autonomous Underwater Vehicle, AUV）的技术日趋成熟并逐渐获得广泛应用，已经在科学考察、海洋资源开发、军事活动等相关领域发挥重要作用。随着人们对AUV能力要求的不断提高，对AUV的使命需求也越为复杂，AUV的自主能力将面临巨大挑战，当前AUV较低的自主性已经成为限制其应用范围的重要原因之一。因此，增强AUV自主能力的相关研究很有必要，也是当前水下机器人领域的研究热点之一。AUV对自身周围环境的感知能力是其根据使命需求进行自主决策的首要前提，AUV需要具有对传感器信息进行融合处理的能力。本文的研究工作依托中国科学院重点项目，在总结国内外目标跟踪相关研究成果的基础上，结合项目对AUV的实际功能需求，对AUV使用主动声呐跟踪机动目标的相关技术进行了深入研究，具体研究内容为：(1) 无偏量测转换方法研究在水下目标跟踪中，主动声呐获得的量测信息为极坐标系下的距离和方位角，而目标的运动方程通常建立在直角坐标系下，坐标系的不相匹配造成量测方程非线性，量测转换方法是量测方程线性化的一种常用方法，但传统量测转换方法在互距离测量误差增大时跟踪性能有所下降。针对这一问题，首先分析了传统量测转换方法的转换偏差，然后对现有去偏量测转换方法的转换偏差和估计偏差进行了分析与比较，在此基础上提出了一种基于量测量和卡尔曼滤波预测量的混合去偏量测转换方法，在跟踪的不同阶段分别基于量测量和卡尔曼滤波预测量，结合无迹变换估计转换量测方差。最后通过仿真验证该转换量测方法较现有方法具有更高的跟踪精度和可信度。(2) 目标运动模型解耦合方法研究应用量测转换方法将目标位置量测信息由球面坐标系转换到笛卡尔坐标系后，转换量测方差矩阵为非对角阵，量测方程于各坐标轴之间耦合，计算量增大。为减少计算量，对目标量测模型的解耦合方法进行研究。首先给出了高维耦合目标的运动模型，然后研究了两类量测方程的解耦合方法，接下来，为去除各坐标轴之间估计结果的相互影响，在正规变换方法的基础上，提出了一种构造修正加权矩阵的改进解耦方法，并详细阐述了修正加权矩阵的构造方法。最后，将该解耦方法与转换量测方法和交互式多模型算法相结合进行了仿真验证，仿真结果表明改进解耦方法可以有效减少计算量，且能够消除各维度估计结果之间的相互影响。(3) 机动目标跟踪自适应滤波算法研究为提高机动目标跟踪定位的精度，对自适应滤波算法进行研究。首先给出了单机动目标跟踪的基本原理，然后对当前较为常用的目标运动模型进行了介绍和仿真分析，接下来，在对机动目标跟踪算法进行简单分类和说明的基础上，提出了一种基于实时辨识自适应滤波的改进交互式多模型算法。该改进交互式多模型算法包含一个基于匀速运动模型的卡尔曼滤波器和一种实时辨识自适应卡尔曼滤波器。后者由两个卡尔曼滤波器组成，目标状态估计滤波器用于对目标的运动要素进行分析，另一个滤波器对目标状态估计滤波器的过程噪声方差进行估计，使过程噪声方差能够根据目标的机动强度进行自适应调整。最后通过仿真验证了算法的跟踪性能良好，能够在一定程度上提高机动目标的跟踪精度。(4) 杂波环境下的数据关联算法研究针对水下目标跟踪环境中存在大量杂波的实际情况，对杂波环境下的单目标数据关联算法进行研究。首先介绍了杂波环境下数据关联的基本原理，然后对几种经典的单目标数据关联算法进行了简要介绍，接下来，在对幅值信息的统计特性进行详细介绍的基础上，考虑到跟踪过程中的目标存在性问题，提出一种基于幅值信息的改进集成概率数据关联算法，将幅值似然比与目标存在概率和量测量关联概率相结合，可以在多杂波环境下改善集成概率数据关联算法的跟踪性能，提高目标存在性判断的快速性，降低目标丢失概率。最后通过仿真验证了改进算法的有效性。(5) 算法联合仿真与实现对基于以上研究内容的机动目标跟踪系统进行软件实现，并设计了一套该软件的辅助验证仿真系统，通过Matlab模拟目标运动轨迹并提供量测信息，同时接收并记录目标跟踪软件计算得到的目标运动位置估计结果，以便于算法性能的分析与比较。最后使用该辅助仿真系统验证了该套软件的可行性。 |
| 索取号 | TP391.41/L35/2015 |
| 英文摘要 | Thanks to the development of the AUV technologies in the last few decades, the AUVs have become mature and gradually widely applied. AUVs have already played an important role in scientific investigation, ocean resources exploitation, military operation, and so on. However, as the requirement of the capabilities of AUVs increases, the missions required for AUVs become more complicated. In consequence, the autonomy of AUVs faces big challenges, whose low level currently becomes one of the important reasons for limiting their scope of application. Thus, relevant researches on promoting the autonomy of AUVs are very necessary, which have become one of the current hot spots in the area of AUV. Surrounding environment perception of AUVs is the first premise of autonomous decision according to mission requirements. Thus, AUVs are required to have the ability of sensor information processing and fusion. This thesis relies on the Key Research Program of the Chinese Academy of Sciences. On the basis of summarizing target tracking related research achievements at home and abroad, combined with the actual function requirement of the program, thorough research and analysis on the maneuvering target tracking by AUVs with active sonar is carried out. Specific contents are as follows:(1) Study of unbiased converted measurements: In underwater target tracking applications, the target position measurements obtained by the active sonar are established in polar coordinates in terms of range and bearing, while the target motion is usually modeled in Cartesian coordinates. The mismatch of coordinates leads to nonlinear measurement equation. The converted measurement is usually used for measurements conversion between defferent coordinates, while the performance of the conventional converted measurement degrades for large cross-range errors. On this issue, firstly, the bias of the conventional converted measurement is analyzed, and then, the converted bias and estimation bias of existed debiased converted measurements are analyzed and compared. On basis of all analysis above, a measurement-conditioned and prediction-conditioned hybrid debiased conversion is presented. The covariance of the converted measurements is approximated by the measurements and the Kalman prediction at different stages of tracking process, combining with Unscented Transform. Finally, simulation results show that the converted measurement Kalman filter proposed is more accurate and credible.(2) Study of decoupled method for target kinematic models: Due to non-zero off-diagonal terms caused by use of converted measurement method for converting the target position information from the spherical coordinate to a Cartesian frame of reference in the covariance matrices of measuring noise, the coupling between different coordinates in measurement equation makes computation greatly increase. In order to reduce the computation burden, the decoupling method for target measurement equation is studied. Firstly, the coupled kinematic state model of targets is given, and then, two kinds of decoupling methods are studied. To eliminate the influence between estimations of each coordinate, on basis of the canonical transform, an improved decoupling method using modified weighted matrix is presented. The design method of the matrix is illustrated in detail. Finally, Simulation combined with unbiased converted measurements and IMM (Interacting Multiple Model) algorithm is carried out and the result indicates the proposed algorithm can reduce computational burden and separate influence among three Cartesian coordinates.(3) Study of adaptive filter algorithm for maneuvering target tracking: In order to improve the tracking accuracy of maneuvering targets, adaptive filter algorithm is studied. Firstly, the basic principle of the single maneuvering target tracking is given. And then, the existed target motion models are studied and analyzed by simulations. On basis of a simple classification of maneuvering target tracking algorithm, an improved IMM algorithm based on a kind of adaptive Kalman filter with real-time identification is given. The improved IMM algorithm contains a stardand Kalman filter and an adaptive Kalman filter with real-time identification. The latter one consists of two Kalman filters. The fliter for target state estimation is used to analyze the target motion, whose process noise variance is estimated online by the other filter to adaptively be adjusted according to the target maneuvering intensity. Finally, simulation results show that the proposed algorithm performs well, which, to a certain extent, effectively improves the tracking accuracy of maneuvering targets.(4) Study of data association algorithm in Clutter: According to the actual situation where there are lots of clutter for underwater target tracking, the study of tracking performance of data association methods for single target is carried out. Firstly, the basic principle of data association in clutter is introduced. And then, Several classic data association algorithm is introduced in brevity. On basis of introduction of statistics of AI in detail, considering the problem of track existence in the estimation process, an improved IPDA (Integrated Probabilistic Data Association) algorithm based on AI is proposed. The likelihood ratio of amplitude is combined with the probability of track existence and the association probability which can improve the performance of estimation process and rapidity of judging existence of targets, and reduce probability of lost track. Finally, simulation results indicate the proposed algorithm validity.(5) Algorithm co-simulation and realization: A maneuvering target tracking system software based on the research contents mentioned above is implemented. An aided verification simulation system for the tracking software is designed using Matlab. The verification system can simulate target trajectory, provide measurement information and record the position estimation of the target from the tracking software to facilitate analysis and comparison for the performance of the algorithm. Finally, simulation results indicate the tracking software validity. |
| 语种 | 中文 |
| 产权排序 | 1 |
| 页码 | 123页 |
| 源URL | [http://ir.sia.ac.cn/handle/173321/16786]  |
| 专题 | 沈阳自动化研究所_水下机器人研究室 |
| 推荐引用方式 GB/T 7714 | 李为. 水下机动目标跟踪关键技术研究[D]. 中国科学院沈阳自动化研究所. 中国科学院沈阳自动化研究所. 2015. |
入库方式： OAI收割
来源：沈阳自动化研究所
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。