无人机智能搜救中的多目标精确定位方法研究

Unmanned aerial vehicles (UAVs) have been widely used in different areas and the related UAV technologies have been developed rapidly in recent years. However, in the search and rescue (SAR) missions, UAVs are not intelligent enough to satisfy the practical requirements. Therefore, it is urgently necessary to develop a fast, accurate and intelligent UAV multi-target localization strategy for SAR missions. In this strategy, the UAV has the intelligent self-decision making abilities of target accurate estimation without any help from a remote human controller. This project focuses on multi-target localization problem using UAV-equipped sensors' target angle-of-arrival (AOA) measurements. The measurements include the processed target angle-of-arrival (AOA) information. Four stages are contained in this project, i.e., measurement noise mitigation, multi-target estimation algorithm, estimation accuracy rapidly promotion and intelligence improvement of the UAV target localization system. First, a measurement noise mitigation algorithm using machine learning methods will be developed to acquire the accurate targets’ positions. Second, the data association problem in multi-target localization is analyzed and a machine learning method will be applied to eliminate the negative impacts caused by data association. Also, the related multi-target estimation algorithms for UAV SAR missions will be proposed. Third, the theoretical relationship between the UAV-target geometry and data association problem, and the mathematical relationship between the UAV-target geometry and estimation accuracy will be found, respectively. Then, a UAV flying path optimization cost function will be designed considering both the data association and estimation accuracy. The appropriate optimization algorithms will be applied to minimize the path optimization cost function. Thus, UAV will be navigated by the results of the flying path self-optimization strategy to improve the localization performance. Finally, the effectiveness and characteristics of the proposed algorithms will be demonstrated and discussed with simulation examples. This project will develop an intelligent UAV multi-target fast and accurate localization method in which the UAV remote human controller is unnecessary. The outcomes will provide a theoretical basis and make a contribution to the practical intelligent UAV SAR applications.

虽然无人机应用科学近年来发展迅速，但是无人机在搜索救援方面相关应用的智能化水平依然难以满足需求。因此，迫切需要一种能够快速实现多目标位置精确、快速确定的智能无人机搜救方法。本项目以如何由无人机机载传感器目标方向角量测中精确提取多目标位置信息为研究问题，在量测噪声抑制、多目标位置实时估计、估计精度快速提升及智能性方面进行深入理论研究。具体内容包括：（1）基于机器学习算法的复杂量测噪声抑制研究，探究相应目标估计理论；（2）分析多目标估计中的数据关联问题，开展基于学习策略的无人机多目标数据关联抑制研究；（3）探究无人机-目标相对位置关系与目标定位精度、数据关联之间的理论联系，构建无人机路径优化代价函数，提出路径优化方法，进一步改进定位速度及精度；（4）最后进行相应计算机仿真实验验证。本项目的完成，将提供一种无需地面遥控的无人机多目标快速精确定位方法，为无人机自主搜索救援的应用提供理论基础。

本项目围绕无人机搜索救援问题中如何快速实现多目标位置精确、快速定位问题，开展相关研究。首先，针对研究内容一“结合机器学习策略的多目标智能估计方法”开展研究。采用超限学习机对目标方位角量测非高斯噪声分布特征开展辨识研究，结合扩展卡尔曼滤波算法实现目标状态精确估计；还采用混合高斯模型结合期望最大化的方法对量测信息中的非高斯噪声进行拟合，提出了通过改善传感器观测位置消除估计偏差的新方法。所提方法通过计算机仿真验证了所提理论的有效性，为目标滤波估计理论做出了贡献。其次，围绕研究内容二“智能路径优化改善定位精度的研究方面”开展研究。（1）针对多目标快速定位估计问题，设计了多目标伪线性卡尔曼滤波算法，并提出了搭载传感器的无人机路径优化策略，通过几何位置推断规避多目标量测数据关联，最终实现局部最优规避而获取目标准确位置估计。（2）探索了传感器位置与目标之间的几何关系与定位精度的联系，提出了针对目标信息传递时间、接收信号强度、目标方位信息以及多种量测混合情况的传感器最优分布策略，并推导了可达的理论最小克拉美罗下界。（3）提出了可以快速改善目标定位精度的无人机移动路径优化方法，以及保证无人机紧密追踪优化轨迹移动的智能追踪控制方法，将学习策略与控制理论结合推导出了新的理论方法。.最终，在本项目的支持下，发表10篇学术论文（6篇SCI，4篇EI），申请发明专利7项，搭建了基于无人机、固定摄像头的验证平台，以安防应用场景中可疑目标的快速定位为实例，开展实验验证，为实际应用奠定了基础。