复杂环境下多传感器随机集目标跟踪方法研究

Because of the low probabilities of detection and recognition, the high false alarm rate and the high target missing rate, it is difficult for multi-target tracking in the complex environments. On the basis of the random finite set (RFS) theory, this project will address the following aspects: 1) The multi-sensor RFS based filters with sum and product and measurement sets partitioning forms will be investigated, in order to solve the scale unbalance and measurement sets partitioning problems and improve the theory for application. 2) For the problems of unknown clutter rate and detection profile, a cardinality modification method is proposed to improve the estimation accuracy of cardinality distribution, and then the target states and the detection profile will be jointly modeled to improve the target state estimation. 3) For the problem of extended and unresolved target tracking, the modeling methods for the measurements of extended targets, the fast partitioning methods based on the fuzzy clustering for extended targets and unresolved targets, and the algorithm for the reduction of Gaussian inverse Wishart mixtures will be researched, respectively. 4) For the track management problem in the RFS based filters, a method based on the multi-scan weighting and fuzzy clustering is proposed, and then some untraditional measurements such as the target attributes and features will be employed to further enhance the performance of track continuity. 5) Many different untraditional measurements will be expressed unitively in the framework of random finite set theory by employing the general likelihood function, and then a conceptual bridge will be constructed between the target recognition model and the target tracking model.

针对低检测率、低识别率、高虚警率、高丢失率等复杂环境下的多传感器多目标跟踪问题，本项目以随机集滤波理论为基础，重点研究：1）基于和积混合和量测集划分形式的多传感器随机集滤波方法，在解决缩放比例失衡和量测集划分问题的基础上，进一步完善该理论，并推动其实用化。2）针对未知杂波密度和检测概率问题，采用联合势分布修正法提高势分布估计精度，并通过对目标状态和检测概率进行联合估计建模，提高滤波精度。3）针对扩展目标和群目标跟踪问题，研究扩展目标的量测建模方法、基于模糊聚类的扩展目标和群目标快速划分方法、高斯逆威沙特混合约简算法。4）针对随机集滤波的航迹管理问题，研究基于多帧加权和模糊聚类的航迹维持算法，并结合目标属性、特征等非传统量测信息提高航迹维持性能。5）通过引入广义似然函数，将各种非传统量测在随机集的框架下进行统一表示和度量，从而在目标识别分类模块和目标跟踪模块之间建立联系的桥梁。

针对低检测率、低识别率、高虚警率、高丢失率等复杂环境下的多传感器多目标跟踪问题，本项目以随机集滤波理论为基础，重点研究复杂环境下的多传感器随机有限集滤波、未知杂波密度和检测概率、扩展目标和群目标跟踪、非传统量测、以及随机有限集航迹管理等关键问题，取得了一系列创新性研究成果，主要包括：（1）针对多传感器随机有限集滤波问题，在线性高斯混合模型下，提出了一种改进的迭代修正高斯混合PHD（IIC-GM-PHD）算法和一种势修正的乘积形式多传感器PHD算法，有效解决了传感器更新顺序对迭代修正多传感器PHD的影响，并提高了目标状态估计的精度；（2）针对未知杂波密度和检测概率问题，提出了一种基于拟蒙特卡罗方法的未知杂波GMP-PHD算法，有效解决了非线性滤波问题；（3）针对扩展目标和群目标跟踪问题，提出了基于模糊ART的量测划分和混合约简方法，从而提高了跟踪系统的效率；（4）针对非传统量测问题，提出了一种IMMBPF算法，极大减少了算法的执行时间；（5）针对随机有限集航迹管理问题，提出了一种基于模糊聚类与多帧加权新的快速数据关联方法，并结合目标属性、特征等非传统量测信息提高了航迹维持的性能。本项目圆满完成了项目申请书既定的研究内容，研究成果在武器装备和防御系统的推广应用方面具有重要的价值。