非完备信息下多源动态异构图像鲁棒融合跟踪方法

For the requirements of continuous and stable targets tracking with incomplete information of moving target state and observation environment, the robust fusion tracking decision strategy for multi-source dynamic heterogeneous image will be studied. And the mathematical model will be proposed for representation for unstructured multi-source heterogeneous information. It is the first time to build such an unified representation for multi-source heterogeneous information with unstructured attributes based on a mathematical model. A heterogeneous information fusion model will be studied based on mapping and association of image and non-image. An approach for multi-sensor visual feature structure based on machine learning dynamic image fusion tracking will be proposed with a multiple attribute decision strategy. The confidence measures and reliability of target tracking system under complex background will be improved.

针对运动目标状态和观测环境非完备信息条件下的目标持续、稳定跟踪需求，围绕联合多属性多源动态异构图像鲁棒融合跟踪决策策略关键科学问题，建立非结构化多源异构信息表征数学模型，实现非结构化属性的多源异构信息统一表征，构建图像与非图像异构信息关联映射及融合模式，提出联合多属性决策策略下基于机器学习的多传感器图像融合跟踪方法，提高复杂背景下目标跟踪系统的可靠性。

针对运动目标状态和观测环境非完备信息条件下的目标持续、稳定跟踪需求，围绕联合多属性多源动态异构图像鲁棒融合跟踪决策策略关键科学问题，提出了无先验信息条件下彩色传递增强技术，提升源图像中目标信息表征能力；提出了基于场景识别、指导滤波、图像统计和四阶偏微分方程的多源同构图像融合方法，提升了图像的品质、信息量和目标可识别性；提出了视觉分类与运动跟踪结合的跟踪算法、基于深度多视角压缩模型的跟踪算法和联合多属性决策的融合跟踪算法，实现了复杂背景非完备信息下基于多传感器图像融合的运动目标快速稳定跟踪，并开展定量性能评价；建立了机载对地半物理仿真平台、无人机跟踪平台和目标跟踪软件平台，开展了对地目标跟踪仿真与实验。