面向禁飞区反无人机的异构多传感器协同探测与跟踪

Unmanned Aerial Vehicle (UAV) is widely used in many industrial areas and mass consumers, but the spaceflight, safeguard, secret and privacy protection are encountering great threats and challenges. In this study, distributed heterogeneous multi-sensors are used to construct a monitoring network with full coverage of the no-fly zone to detect, identify and track the illegal invasion UAV in time. The main research issues include: (1) the self-organizing management and collaborative scheduling strategy of distributed heterogeneous multi-sensors, visible/infrared images fusion guided by target energy entropy, and collaborative detection of dim small target by using low-rank and sparse matrix decomposition; (2) the construction of polymorphous sparse coding library about UAV, identification of UAV type through multi-source information fusion, and visual positioning for UAV by utilizing distributed multi-sensors; (3) pose estimation of UAV by using its three-dimensional model projection in different directions, and multi-sensor alternate tracking for mobile UAV under complex background combined with temporal-spatial clues of different maneuver. This study is closely cater for the current urgent needs of anti-UAV, and aims to research key technologies for a flexible deployment, low cost and intelligent efficient solutions. Furthermore, the research achievements will provide technical support and theoretical guarantee for engineering application and have great social benefits and promising application.

无人机被许多工业领域和广大消费者广泛使用的同时，也给航空飞行、安全保卫、泄密防护、隐私保护等带来了极大的威胁和挑战。本课题采用分布式异构多传感器构建对禁飞区空域全覆盖的监控网络，及时探测、识别与跟踪非法侵入禁飞区的无人机。主要研究内容有：(1)设计分布式异构多传感器的自组织管理与协同调度策略，以目标能量熵为引导融合可见光/红外图像，采用低秩-稀疏矩阵分解技术实现弱小目标的协同检测；(2)构建无人机多形态特征稀疏编码库，通过多源信息融合识别无人机类型，并利用分布式多传感器对无人机进行视觉定位；(3)利用无人机三维模型在不同方向上的投影，估计无人机位姿，并结合不同机动动作对应的时空域线索，实现对复杂多变背景下机动无人机的多传感器交替跟踪。本课题紧扣当前社会迫切的反无人机需求，研究灵活布设、低成本、智能高效的反无人机关键技术，为工程应用提供技术支持和理论保障，具有巨大的社会效益和广阔的应用前景。

本项目针对无人机在各行业被广泛应用的同时给航空飞行、安全防卫等带来威胁与挑战这一问题，在调研与分析国内外反无人机技术现状的基础上，围绕利用分布式异构多传感器构建对禁飞区空域全覆盖监控网络所涉及的异构多传感器自组织管理与协同调度、低空空域飞行小目标检测、多源图像中的目标分类识别、无人机目标空间定位与鲁棒跟踪等环节的关键技术开展研究，取得了一定的研究进展与成果。首先，通过研究分布式传感器网络的自组织分簇路由与定向扩散路由，实现对分布式部署探测器的自组织管理与协同调度，从而对禁飞区空域全覆盖监控；然后，通过研究动态场景红外图像背景建模、基于压缩感知的全局/局部异常事件检测、基于高斯混合模型的压缩感知域飞行小目标检测、基于生成式低秩背景估计的飞行小目标序列化检测，实现对进入监控范围的无人机等空中飞行目标的及时探测；接着，通过研究基于区域集成学习网络的目标细粒度分类、基于深度多层融合密集网络的高光谱图像目标分类、融合空洞卷积神经网络的语义SLAM，实现对多源图像中目标的分类识别；进而，通过研究卫星导航选星优化与完好性监测、低空无人机飞行障碍物冲突检测，以及基于ADS-B的合作无人机目标自动跟踪，实现对地面分布式探测器节点的辅助定位与空中无人机的相对定位，进而实现对空中飞行目标的鲁棒跟踪；最后，搭建了集无人机飞行、数据采集与视觉探测、服务器算法处理及云台控制跟踪于一体的反无人机演示系统。本项目研究成果对于构建灵活布设、低成本、智能高效的反无人机监控网络提供了理论支撑，可以被广泛应用于机场净空区、政府部门、军事禁区、核电站、医院、学校等场所，对推动无人机产业健康快速发展，保障人民生命财产安全，具有巨大的经济效益和社会效益。