针对目标检测跟踪问题的贝叶斯非参建模方法研究

项目名称： 针对目标检测跟踪问题的贝叶斯非参建模方法研究

项目编号： No.61302158

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 无线电电子学、电信技术

项目作者： 刘斌

作者单位： 南京邮电大学

项目金额： 24万元

中文摘要： 常规目标检测跟踪方法通常基于相邻检测间目标接收数据统计平稳性假设,采用特定参数形式的目标信号与目标运动模型，模型失配造成系统整体性能无法满足实际需求。本项目研究贝叶斯非参建模方法，对目标检测跟踪系统中可能影响接收信号统计特性的随机因素进行统一建模，从系统角度为目标检测与跟踪问题提供灵活、准确的模型解释和全局最优解。具体地，本项目主要研究：1)对于可能影响观测信号统计特性的随机因素，对其先验知识进行贝叶斯非参建模的方法；2）由观测数据引导的自适应的模型构建与更新机制；3）相应的贝叶斯蒙特卡洛采样推理算法。本项目所提供的建模计算方法最大限度地挖掘与利用先验信息，实现先验信息与观测数据信息的最优融合，将显著提高目标检测增益与目标跟踪精度。

中文关键词： 信号检测；目标跟踪；贝叶斯建模与计算；信息融合；鲁棒方法

英文摘要： Traditional target detection and tracking methods were usually developed based on the assumption of statistical stability of the data received from adjacent detections. So most of them utilized ad hoc parameterized target signal and movement models. The model mismatch deteriates the global system performance, which is not able to meet the real demands. This project aims to develop a suite of Bayesian nonprametric modeling tools to cover the possible random factors that may affect the statistical properties of the observational data in a unitary model, providing more flexible and precise model interpretations and the global optimal solution. Specifically this project mainly researches on: 1)Bayesian nonparametric priors for describing prior knowledge for factors which may influence the statistical properties of the receiving data; 2) Adaptive mechanisms for model construction and updating guided by observational data; 3) Corresponding Bayesian Monte Carlo sampling and inference algorithms. The proposed methods explore and utilize the prior knowledge to a maximum extent and implement the optimal fusion of prior knowledge with the observational data. It's likely to be capable of remarkably improving the target detection gain and tracking precision.

英文关键词： signal detection；target tracking；Bayesian modeling and computation；information fusion；robust methods