无人机自主导航中LiDAR点云与图像特征提取与配准方法研究

Unmanned Aerial Vehicle(UAV) autonomous navigation technology has become an important research direction in the field of navigation and have great potential application in military and civilian areas. Systems using satelliates to aid INS suffer the problem of signal blocking and jamming.So independent integrated navigation system, which sensing the environment to aid Inertial Navigation System(INS), are regarded as one potential effective solution. Compare with many kinds of senors, vision camera and Light Detection and Ranging(LIDAR) are able to mapping 3D environment directly and accurately. So it is possible to integrate LiDAR and Vision and for precise navigation on military or civil area. . In this proposal, the consistency of Vision and LIDAR integrated multiple-source localization in UAV autonomous navigation system will be considered, including consistency problem for feateure detection, measurement consistency problem among Vision and LIDAR system. The problems are tight parts for a whole.. This proposal will discuss local interest point detector algorithm in point cloud with multi-scale consistency. In this work, the multi-scale point cloud local feature extraction algorithm will be in depth research by analyzing the point cloud smoothing algorithm with the establishment of the multi-scale point cloud.. Then,the visual features and point cloud features will be registrated to improve the localization and navigation results. A cascade structure of fusing visual features and LIDAR point cloud features will be proposed.. All in all, camera and LIDAR are both popular senors in localization and autonomous navigation application, it is worthful to research on registration of UAV localization and autonomous navigation using Vision and LIDAR integrated system.

小型无人机平台飞行稳定性低，获取的LiDAR和图像数据质量往往较差且存在尺度变化，导致LiDAR点云与图像数据的鲁棒、快速配准依然是一个难题，也是目前是制约基于LiDAR和图像传感器的组合导航技术在无人机自主导航领域广泛应用的关键问题。针对此问题，本课题研究适于高动态平台自主导航的LiDAR点云与图像特征提取与配准方法。首先研究多尺度一致的点云提取算法，建立多尺度点云，并通过特征曲率检测方法提取不同尺度下的几何局部特征，在多尺度几合局部特征基础上研究相应特征描述符和匹配方法，实现多尺度一致的点云局部特征提取；然后，在多尺度一致点云局部特征提取研究基础上，开展基于边缘片段和控制点的局部图像特征与LiDAR点云局部特征的匹配方法，并对其中涉及的关键问题进行实验验证；最终，提出适于高动态平台、小型化LiDAR和图像传感器组合导航特点的特征提取与配准方法。

本项目以无人机自主导航应用需求为牵引，围绕LiDAR点云与图像特征提取与配准问题中的基础理论与关键技术开展研究，取得了重要的进展和成果。.1.针对多尺度一致的局部特征提取问题，项目基于单幅图像的深度估计算法，提出了融合深度信息和几何结构信息两种特征的特征提取与匹配方法，取得了更好的提取与匹配结果。此部分研究成果发表在了业界高水平SCI期刊（影响因子3.558）《IEEE Transactions on Circuits and Systems for Video Technology》上。.2.针对LiDAR点云数据与图像数据配准问题，项目提出了基于三维最小生成树的立体匹配算法，设计了多视角三维标签的聚合代价，取得了比极线几何约束的局部块匹配更精确的结果。该算法发表在了SCI期刊《APPLIED OPTICS》上（影响因子1.791），并在权威立体匹配算法排行榜Middlebury 3.0上持续8个月排名第一。.3.在无人机自主导航方面，项目基于前两项研究成果，拓展研究了无人机自主导航与避障算法。通过深入分析激光雷达精度与距离的特性关系，研究了一种分层导航与障碍规避方法，提出了一种克服V 型障碍陷阱的激光雷达机器人分层避障方法。为模糊规避区设计了神经网络障碍规避算法，同时在精确规避区采用边界点追踪法避撞，相关研究成果发表在核心期刊《控制与决策》和EI会议上。.4.为了验证所提出的LiDAR点云与图像特征提取与配准算法，项目基于旋翼无人机、移动平台、线阵激光雷达、立体摄像头及便携式电脑搭建实验验证平台，成功验证了所提出算法的可行性和精度性能。.截止目前，课题组已发表学术论文7篇，在投2篇，其中在APPLIED OPTICS、IEEE Access、IEEE Transactions on Circuits and Systems for Video Technology等SCI期刊发表论文4篇，其他核心期刊、EI会议论文3篇，标注率100%。出版译著一部，申报专利1项（已提交申请）。培养研究生6名，项目组通过积极参与相关领域主流国际会议、邀请国外专家访问等方式，与国外同行进行了深入的学术交流与合作，为达到小型无人机在无GPS环境下的自主导航提供理论基础和技术支撑。