扩展目标跟踪算法及其性能评估的研究

With the increasing of the sensor resolution capability, a target gives rise to several measurements from different measurement sources, and thus the assumption which a target is considered as the point target in the most tracking application, is no longer valid. The extended target tracking has become an urgent research subject in the military field. In this project, tracking algorithms and its performance evaluation of the irregular shape extended target will be deeply researched. The research mainly includes the following aspects: Firstly, the multiple sensor tracking algorithm for the irregular extended target tracking of nonlinear system will be researched, and the relationship between the measurement distribution and the sensor parameters and the relative geometric position of the target-sensor will also be studied on; Secondly, to track multiple extended targets for the nonlinear system in the presence of clutter measurements, the target state and data association will be jointly estimated; Thirdly, the performance evaluation for the irregular shape, and the comprehensive performance evaluation for the kinematic state and the shape, will be discussed; And then, the target state estimation lower bound, which could provide a basis for the performance evaluation of the filtering algorithm and for the extended target model optimization, will be researched; Fourthly, to enhance the tracking accuracy and the class recognition accuracy, the scheme of the information exchange between the tracking module and the class recognition module will be researched, and the joint target tracking and class recognition algorithm will be proposed; Finally, the performance of algorithms will be tested and verified through simulations and experiments. This project will improve the ability of the target tracking and the class recognition in military or civil filed, and has important practical significance to enhance the defense and attack abilities for our military system.

随着传感器分辨率的提高，一个目标在每个时刻可产生多个量测，点目标假设不再成立，扩展目标跟踪成为军事领域亟需研究的课题。本项目拟对扩展目标跟踪算法及性能评估进行深入研究，主要内容有：研究非线性系统、不规则扩展目标的多传感器跟踪算法，并研究量测分布与传感器参数、目标-传感器的几何位置之间的关系；采用联合估计目标状态和数据关联的策略，解决杂波环境下非线性系统的多扩展目标跟踪问题；研究不规则形状估计的性能评价指标，及目标运动状态和形状估计的综合性能评价体系；研究目标估计误差下界，为滤波算法评估、扩展目标模型的合理性评价和优化提供依据；研究目标跟踪和类型识别模块间的信息交换方案，提出扩展目标联合跟踪和类型识别算法，提高目标跟踪精度和类型识别的准确率；最后，通过仿真和实验对算法性能进行测试和验证。本项目研究将提高军事或民用领域的目标跟踪和类型识别能力，对提高我国军事防御和攻击能力具有重要的现实意义。

本项目针对扩展目标跟踪中的问题进行了研究，主要研究内容如下：.（1）针对形状估计较差、无法体现形状主要特征的问题，使用水平集隐含地更加精确地表示扩展目标形状，使用高斯曲面拟合的方式建立包含形状信息的量测源分布模型，提出了水平集阈值和高斯曲面协方差的自适应算法，实现了较为精确的扩展目标形状估计。.（2）针对采用多个时刻量测集拟合量测源分布，导致计算量随时间增加急剧增加的问题，实现了在Bayes框架下形状的递推估计，并采用滤波精度较高的容积卡尔曼滤波器，对扩展目标的运动状态进行估计，进而实现了扩展目标的跟踪。.（3）针对非线性非高斯的多扩展目标跟踪，将粒子滤波技术应用到多扩展目标的运动状态和形状估计中，实现了数据关联和目标状态的联合估计，避免了多扩展目标跟踪中的量测集划分问题。实现了当目标演化及量测模型退化为线性高斯系统时问题的简化处理，用粒子滤波对数据关联进行估计，采用适用于线性高斯的扩展目标滤波器解决目标状态和形状的估计问题。.（4）针对粒子滤波存在的粒子贫化现象导致跟踪精度下降的问题，将粒子状态值看作麻雀个体位置，通过麻雀搜索算法的位置更新机制代替粒子滤波的重采样引导粒子向高似然区域移动，将粒子滤波的状态估计转变成麻雀种群觅食寻优。分析麻雀搜索算法应用于粒子滤波时存在的问题，并进行相应的改进，使均方根误差更小，粒子的分布更合理。.（5）针对现有扩展目标形状估计评价指标无法较好地度量形状特征的相似程度的问题，提出了采用径向比值的标准差描述估计形状和真实形状的相似程度，采用径向比值的均值描述两个形状的大小关系，能够较好的度量特征相似和大小差异。.（6）针对扩展目标和识别进行联合识别的问题，建立了目标类型与目标运动状态和形状演化模型之间的关系，并根据目标运动状态和形状的估计建立判断目标类型的方法，基于联合概率密度-质量函数实现对扩展目标跟踪和类型识别之间信息的交换，采用近似算法在牺牲较小代价的基础上减少计算量。

内容获取失败，请点击重试