基于变分贝叶斯的分布式多传感器多目标跟踪建模理论研究

Study on the distributed multi-sensor multi-target tracking plays an important role in national defense and civil security. However, theory of modeling the tracking is still not sufficiently exploited at present. Systemic theory and analysis are absent as well. Combining nonparametric Bayesian theory, optimization theory, signal processing and other fields of knowledge, multiple maneuvering targets tracking where the number of targets is time-varying on the distributed multi-sensor system is mainly discussed in this project. Especially modeling the tracking system is posed as the mainline of the proposed project. In order to improve tracking performance, the proposed project considers three key factors, which are track registration, track association and track fusion, in a joint optimization framework. Based on the variational Bayesian algorithm, distributed multi-sensor multi-target tracking theory is systematically studied and the correlation of these three key factors is revealed. Moreover, the integration mechanisms of these three key factors for improving tacking performance are further explored. Ultimately, a system of modeling strategy and corresponding solving method for distributed multi-sensor multi-target tracking is proposed. Therefore, the proposed project will provide a new thought and method for the conversion from the "dispersion modeling" to "integrated model" in an unknown complex environment, and then offer theoretical and technical support to improve the performance of the distributed multi-sensor multi-target tracking.

分布式多传感器多目标跟踪的研究对国防和民用安全具有极其重要的意义。 但目前其建模理论的研究还很不够，缺乏一套系统的理论和方法。本项目结合非参贝叶斯理论、最优化理论和信号处理等学科知识，在分布式多传感器系统下，以数目不定的机动多目标跟踪为研究对象，以跟踪建模为主线，以提高跟踪性能为目标, 以航迹配准、航迹关联和航迹融合这三个影响跟踪性能的关键因素的联合建模为主要思想, 以变分贝叶斯方法为技术手段，系统研究分布式多传感器多目标跟踪机理，揭示航迹配准、航迹关联和航迹融合这三个关键因素之间的相互关系，探索提高跟踪性能的"多关键因素"的联合机制，提出一套跟踪建模技术和求解算法，为实现在复杂未知环境下的"分散建模"到"一体化建模"的转变提供全新的研究思路和方法，进而为提高分布式多传感器多目标跟踪性能提供理论和技术支撑。

分布式多传感器多目标跟踪的研究对国防和民用安全具有极其重要的意义。航迹配准、航迹关联和航迹融合是分布式多传感器多目标跟踪三个最为重要的部分。但这三个部分互相耦合，相互关联，本项目提出了一套航迹配准、航迹关联和航迹融合的联合模型及其求解算法。取得的理论成果如下：..1. 提出了基于在线贝叶斯方法的联合配准和融合的算法。该算法分析了配准和融合的关系，构建了面向在线估计的配准和融合的一体化优化函数，从而建立了联合模型及其求解算法，最后仿真结果表明此方法比先配准后融合的方法有较好的估计性能。..2. 提出了多源图像配准和融合的联合模型及其求解算法。面向多源图像多目标跟踪问题中的多图像配准和融合问题，首先提出了多图像集成配准算法，然后又提出了多图像集成配准和融合的联合模型及其算法，最后仿真结果表明此方法比传统的先配准后融合的方法有较好的估计性能。..3. 提出了航迹配准、航迹关联和航迹融合的联合模型及其求解算法，并应用到多源图像多目标跟踪上。构建了基于最大似然估计的航迹配准、航迹关联和航迹融合的联合优化函数，并利用最大期望算法进行模型求解，最后仿真结果表明此方法比先配准，然后关联，最后融合的方法有较好的估计性能。