【作者】 杨柏胜；

【导师】 姬红兵；

【作者基本信息】 西安电子科技大学 ， 模式识别与智能系统， 2009， 博士

【摘要】 被动多传感器目标跟踪是多传感器数据融合的一个重要研究内容,在军用和民用领域具有广阔的应用前景,备受国内外学者和工程领域专家的关注。本文针对被动多传感器探测的目标跟踪问题,从系统参数优化设计和目标跟踪方法等方面进行了深入、系统的研究,提出了一些有效的新方法。取得主要成果如下：1在系统参数优化设计方面,提出了一种基于目标跟踪精度分析的系统参数优化设计算法,采用多传感器集中式融合方式与扩展卡尔曼滤波(EKF)相结合实现被动目标跟踪,并推导了跟踪误差的克拉美-罗下限。在此基础上,给出了监视空域内目标跟踪精度的几何分布(GDTE)。此外,通过分析系统参数对目标跟踪精度几何分布的影响,给出了提高系统性能的有效措施。2在单目标跟踪方面,提出了一种基于无迹卡尔曼滤波(UKF)的被动目标跟踪算法,将无迹变换(UT)引入卡尔曼滤波,避免了传统的扩展卡尔曼滤波的线性化近似过程,在保证目标跟踪实时性的前提下,有效提高了目标跟踪精度。针对单一坐标系下滤波算法中目标状态耦合问题,提出了一种基于混合坐标系无迹卡尔曼滤波(HC-UKF)的被动目标跟踪算法,利用无迹变换实现目标状态的坐标转换,降低了目标状态的耦合程度,有效提高了目标跟踪精度。针对基于粒子滤波(PF)的被动目标跟踪算法计算量大的问题,提出了一种基于拟蒙特卡罗采样高斯粒子滤波(QMC-GPF)的被动目标跟踪算法,通过拟蒙特卡罗采样实现高斯粒子滤波的递归计算,减少了样本点数量,在保证目标跟踪精度的前提下,有效降低了算法的计算复杂度。3在机动目标跟踪方面,提出了一种基于二次加权变结构多模型(RVSIMM)的被动机动目标跟踪算法,将二次加权过程引入到模型交互过程中,提高了模型融合精度。为了避免不匹配模型的影响,引入模型集合自适应调整过程,在降低计算复杂度的同时,提高了目标跟踪精度。提出了一种基于自适应两阶段扩展卡尔曼滤波(RTSEKF)的被动机动目标跟踪算法,采用两阶段卡尔曼滤波分别估计目标状态和机动偏差扰动,通过对机动偏差的在线估计来修正滤波输出,并引入改进的噪声自适应估计算法实时估计模型噪声参数,在不明显增加计算量的情况下,有效提高了目标跟踪精度。4在多目标跟踪方面,针对杂波环境下数目未知且时变的多个机动目标的跟踪问题,提出了一种基于交互多模型概率假设密度(IMM-PHD)滤波的被动机动多目标跟踪算法,将多目标状态和观测建模为随机有限集合,并通过概率假设密度滤波(PHD)同时估计目标数目和状态。为了避免目标机动时出现失跟现象,将交互多模型算法(IMM)引入到滤波递归过程中,有效提高了目标跟踪精度。更多还原

【Abstract】 The techniques of passive target tracking are important topics of the research on multi-sensor data fusion. Because of their wide applications in both military and civil areas, much attention has been paid to their developments by worldwide researchers and engineers. Aiming at the techniques of target tracking based on passive multi-sensor detection, this dissertation mainly involves some significant aspects, such as the optimal design of system parameters, target tracking algorithms etc. Some novel efficient methods have been proposed. The main contributions of the dissertation are as follows:1 For the problem of the optimal design of system parameters, an optimal design method based on the analysis of target tracking accuracies is proposed, which involves passive target tracking achieved by introducing extended Kalman filter (EKF) into the multiple-senor centralized fusion scheme, and the Cramer-Rao lower bounds of tracking errors are deducted. Furthermore, the geometrical dilution tracking error (GDTE) in the surveillance area is given. Some efficient approaches to improve system performance are brought forward on the basis of the analysis of the influences of the system parameters to GDTE.2 For the problem of single target tracking based on noisy passive measurements, an unscented Kalman filter (UKF) based passive target tracking algorithm is proposed, in which the unscented transformation (UT) is introduced into Kalman filter with the scheme of multi-senor centralized fusion, by which the numerical accuracies are enhanced with little additional computational costs because of the avoidance of errors from the linearization in extended Kalman filter (EKF). To solve the problem of state coupling in single coordinate based filters, a hybrid coordinates UKF (HC-UKF) based passive target tracking algorithm is proposed, which emploies UT to achieve the nonlinear coordinate transformation. The numerical accuracies of target tracking are improved due to the depressed target’s state coupling. Furthermore, a quasi Monte Carlo sampling Gaussian particle filter (QMC-GPF) based passive target tracking approach is proposed to solve the problem of heavy computational costs for particle filter (PF) based methods. The algorithm introduces quasi Monte Carlo technique into Gaussian particle filter recursion, by which the computational costs are reduced because of the reduced number of samples, and the numerical accuracies are enhanced.3 To solve the problem of maneuvering target tracking, a reweighted variable structure interacting multiple model (RVSIMM) algorithm is proposed. The method introduces reweighted steps into the interacting of multiple models. Moreover, the model sets in each recursion is modified to avoid the blights from the invalid models. Consequently, both of computational costs and numerical accuracies are improved. Furthermore, a robust two stage EKF (RTSEKF) for passive maneuvering target tracking is proposed. The algorithm estimates both the target’s state and its maneuver bias input online, where the estimation of target’s state is modified by the estimation of the maneuver bias. At the same time, the noise statistics parameters are estimated to revise the models of filtering. Consequently, the numerical accuracies are enhanced without too much additional computational costs.4 For the problem of passive multi-sensor multi-target tracking, when the time varying number of targets maneuver and the measurements are mixed with the interferences of clutters, the methods based on random finite set (RFS) are addressed. Then, an IMM probability hypothesis density (IMM-PHD) filter based passive multiple maneuvering targets tracking approach is proposed. The algorithm involves modeling both of the time varying number of targets’states and passive measurements as RFSs. Besides, the interacting multiple model (IMM) is embedded into the filtering recursions to avoid the loss of maneuvering targets. Consequently, tracking accuracies are improved.更多还原