Research On UAV Mapping Method And Application Based On SLAM

UAV(Unmanned Aerial Vehicle)mapping is an important means of 3D reconstruction of outdoor large-scale scenes,and is widely used in land surveying and mapping,engineering quality inspection,historical heritage protection and other industries.At present,the mainstream UAV mapping solutions include oblique photogrammetry and airborne Li DAR(Light Detection And Ranging)mapping.However,the precision of oblique photogrammetry is poor and the reconstruction speed is slow.Airborne Li DAR mapping is costly and requires accurate GNSS(Global Navigation Satellite System)information.To address the above problems,this paper builds a UAV platform equipped with multiple sensors and studies a SLAM(Simultaneous Localization And Mapping)based UAV mapping method.The main work is as follows:(1)For the needs of UAV mapping applications,this paper builds a UAV platform equipped with IMU(Inertial Measurement Unit),binocular fisheye camera and solidstate Li DAR,and calibrates each sensor with internal and external parameters.(2)For the problems of small field of view and irregular scanning of solid-state Li DAR,this paper proposes an adaptive GICP(Generalized-ICP)algorithm based on PCA(Principal Component Analysis)for point cloud registration.The algorithm extends the face-to-face GICP registration to point-to-point,point-to-face,and faceto-face by PCA analysis,and is able to switch between modes adaptively.The algorithm makes full use of the structural and non-structural information in the scene and has high operational efficiency.It is demonstrated that the algorithm has better accuracy and robustness for solid-state Li DAR point cloud registration compared with commonly used Li DAR point cloud registration algorithms.(3)For the problems of IMU integration drift,low vision localization accuracy,and easy degradation of solid-state Li DAR point cloud registration,this paper proposes a multi-sensor fusion SLAM algorithm.The algorithm jointly optimizes the pre-integration constraint constructed by IMU,the reprojection error constraint constructed by image,and the point cloud alignment constraint constructed by Li DAR using LM(Levenberg-Marquardt)algorithm,and is able to output high-precision poses and maps.Through indoor and outdoor experiments,it is proved that the algorithm proposed in this paper has accurate positioning ability and high-quality modeling ability.This paper also shows the application of the point cloud model generated by this system in the field of architecture,showing the practical application value of this system.