【作者】 李良群；

【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， 2007， 博士

【摘要】 目标跟踪与数据关联技术是信息融合系统研究的一个重要课题,由于其在军事和民用领域已经展现出有效而广阔的理论和应用前景,备受国内外学者和众多工程领域专家的高度关注。本论文针对信息融合系统中的目标跟踪与数据关联技术,从单传感器的实时目标跟踪、多被动传感器的机动目标跟踪、红外弱小目标检测前跟踪、多传感器多目标跟踪等几个方面进行了深入系统的研究,提出了一些实时有效的新方法。各章内容安排如下:第一章简要介绍了本文研究的背景、意义、信息融合系统及多传感器目标跟踪系统,概述了当前信息融合、目标跟踪及检测前跟踪技术的研究现状,最后给出了本文的主要研究成果和全文的内容安排。第二章介绍了目标跟踪的基本理论及其数学描述。对现有的多目标跟踪方法进行了综述,以简明表格的形式对几十种不同的多目标跟踪方法进行了分类,并对算法的性能指标进行了评估对比。第三章针对杂波环境中目标跟踪的实时性问题,提出了一类快速实时数据关联新方法,包括最大熵模糊概率数据关联滤波器(MEF-PDAF)和最大熵模糊联合概率数据关联滤波器(MEF-JPDAF)。为了提高目标跟踪的实时性,提出了一种用模糊聚类隶属度代替目标关联概率的权值分配新方案,并根据最大熵模糊聚类的特点,定义最大有效距离剔除大量无效观测,从而减少了计算量。第四章在第三章的基础上,针对杂波环境中被动机动目标跟踪的实时性问题,提出了交互多模型最大熵模糊概率数据关联滤波器,建立了被动多传感器的观测模型,给出了算法的结构流程,并在不同仿真条件下对算法性能进行了验证。为了提高算法的跟踪性能,将粒子滤波与交互多模型结合,提出了一种适合于被动传感器系统的机动目标跟踪算法,利用粒子滤波对非线性问题处理的优势,推导了杂波环境中粒子滤波似然函数的表达形式,给出了算法的结构流程,并对算法性能进行了评估比较。第五章在第四章基础上,继续对粒子滤波及检测前跟踪算法作了进一步研究。提出了一种基于迭代扩展卡尔曼滤波的粒子滤波新方法,利用迭代扩展卡尔曼滤波产生粒子滤波的重要性密度函数,使重要性密度函数能够融入最新观测信息的同时,更加符合真实状态的后验概率分布。在此基础上,提出了一种基于粒子滤波的红外弱小目标检测前跟踪算法(TBD),给出了算法的流程,并用真实红外图像对算法进行了仿真验证。第六章在FCM数据关联的基础上,提出了一种基于多FCM数据关联的多目标跟踪算法。为了使算法适应于杂波情况,提出了一种改进的FCM数据关联更多还原

【Abstract】 The techniques of target tracking and data association are important topics of the research on information fusion system. Because they have been found wide applications in both military and civil areas, many countries and researchers have paid much attention to the development of the target tracking and data association technology. According to the technologies of target tracking and data association in information fusion, this dissertation mainly involves four aspects: the real-time target tracking of single sensor, multi-sensor multi-target tracking, maneuvering target tracking with multiple passive sensors, and track-before-detect of infrared small target. A series of new methods for engineering applications have been proposed. The main contents of the dissertation are as follows:In Chapter 1, the research background and significance of this dissertation, information fusion system and multi-sensor multi-target system are briefly described. The current research of information fusion,target tracking and track before detect are summarized. Finally, the main achievement and arrangement of this dissertation are concluded.In Chapter 2, the basic theory and mathematics derivation of target tracking problem are described. A summary of the techniques of the multi-target tracking methods appeared in both domestic and foreign publications is provided, and the techniques have been categorized into more than 62 different algorithmic types. At the same time, a comparison of the main performance of the algorithms is analyzed.In Chapter 3, for real-time target tracking in clutter environment, a category novel fast data association method is proposed; including the Maximum Entropy Fuzzy Probabilistic Data Association Filter (MEF-PDAF) and the Maximum Entropy Fuzzy Joint Probabilistic Data Association Filter (MEF-JPDAF). In order to improve the real-time of target tracking, a new weight assignment that the joint association probability is replaced by utilizing the fuzzy membership degree of the target and the measurement is proposed. According to the characteristic maximum entropy fuzzy clustering, the maximum validate distance is defined, which enables the algorithm eliminate those invalidate measurements and reduce the computational load.In Chapter 4, for maneuvering target tracking with multiple passive sensors in clutter environment, a novel interactive multiple model maximum entropy fuzzy probabilistic data association filter is proposed. The structure flow chart of the algorithm is given, and a nonlinear measurement model of multiple passive sensors is更多还原