基于广义贝叶斯准则的复杂环境下地面目标联合跟踪与分类研究

In recent years, the battlefield environment becomes more and more complex, the traffic condition continuous to deteriorate, the sensor technology changes rapidly, and these result in the increasing demand of ground target information processing. The joint tracking and classification of ground targets in complex environment is the core problem in this area. The key difficulty of this problem lies in that tracking and classification are completely different but highly coupled. In the existing methods, tracking and classification are handled separately, and as a result their performances are greatly limited. This project handles the above two problems jointly, and finally achieves the best joint performance of tracking and classification. First, considering the complex characteristics of ground targets, i.e., strong road constraints, complicated target motion, high clutter density etc., and under the premise of utilizing the coupling between tracking and classification, this project does research on systematic modeling of ground target tracking and classification; Second, this project sets the integrated framework of tracking and classification. By utilizing the coupling between tracking and classification, their respective performances could be mutually improved. Specifically, the generalized Bayes joint risk function is proposed, the optimal joint decision and estimation solution is also obtained, the balance mechanism of tracking and classification is explored, and the joint performance evaluation metric is designed. Finally, by feature extraction of the characteristics of the multi-source heterogeneous data, this project proposes the multi-sensor data-based ground target joint tracking and classification method. Aforementioned research will provide a theoretical basis for solving the mixed optimization problem of estimation-decision, and further develop the ground target information processing under complex environment, and thus have both theoretical value and practical significance.

近年来，地面战场环境日趋复杂，道路交通状况持续恶化，传感器技术日新月异，使得对地面目标信息处理的需求日益高涨。复杂环境下的地面目标联合跟踪与分类作为其核心难题，本质难点在于跟踪和分类性质不同却高度耦合。现有研究将二者分而治之，导致性能大幅受限。本项目拟一体化解决上述问题，达到联合最优的跟踪和分类性能。首先，针对地面目标道路约束强、运动模式多、杂波密度高等复杂特点，在兼顾耦合的前提下，实现跟踪与分类的系统性建模；其次，建立地面目标跟踪与分类的一体化解决方案，充分利用二者的耦合，实现性能的相互提升。包括：提出广义贝叶斯联合风险，研究最优解，探究跟踪与分类的平衡机制，设计联合性能评估指标；最后，基于多源异构数据特征提取，提出多传感器数据地面目标联合跟踪与分类。本项目的研究成果将为估计-决策混合优化问题的求解提供理论依据，并进一步发展复杂环境下的地面目标信息处理，具有重要的理论价值和实际意义。

近年来，地面战场环境日趋复杂，道路交通状况持续恶化，传感器技术日新月异，使得对地面目标信息处理的需求日益高涨。复杂环境下的地面目标联合跟踪与分类作为其核心难题，本质难点在于跟踪和分类性质不同却高度耦合。鉴于此，本项目一体化解决了上述问题，以获得最优的跟踪和分类整体性能。首先，提出了复杂道路网约束下的地面目标联合跟踪与识别一体化解决方案，解决了各种约束条件下的目标联合状态估计及类型识别，为后续其他复杂条件下的跟踪与分类提供了基础；其次，针对实际复杂运动场景，提出了复杂运动模式（如高机动性、高度非线性等）下的地面目标联合跟踪与分类方法，该方法通过引入基于无迹卡尔曼滤波和期望模式扩增变结构多模型方法，并利用跟踪和识别之间的耦合性，最终提高了其整体联合性能；再次，针对多目标场景，并结合多目标本身的复杂性和特殊性，并充分考虑目标个数推断和状态估计之间的耦合信息，提出了性能优越且易于实现的多目标联合检测与跟踪解决方案；最后，针对扩展目标并结合实际需求，通过引入内嵌式分类，建立了基于随机矩阵的多模型扩展目标跟踪方法，所提出的含有内嵌式软分类和硬分类的扩展目标跟踪方法，不仅满足了实际需求，而且充分利用了分类对跟踪的辅助。上述研究都深入挖掘了不同条件下目标跟踪与分类之间的本质耦合关系并加以有效利用，从而大大提高了跟踪与分类的整体联合性能。本项目的研究成果将为涉及估计决策混合优化问题的求解提供重要的理论支撑，同时对于实际复杂环境下的地面目标信息处理具有重要的理论指导意义和实际应用价值。