Research On Multi-source Data Fusion Technology For Near-field Array 3D SAR

Near-field array 3D SAR imaging technology can obtain 3D SAR images reflecting the electromagnetic scattering characteristics of the detection scene.In recent years,it has been gradually used in the field of target scattering measurement and concealed target detection.However,there are some problems existing in near-field three-dimensional array SAR images.(1)Interference and noise make it difficult to distinguish the target and background;(2)High sidelobe and missing target shape lead to low localization accuracy of target scattering characteristics;(3)Grayscale image is not conducive to target category judgment and scene understanding.These problems will seriously affect the scattering characteristic diagnosis,target detection and target recognition.Therefore,in this thesis,Lidar and cmaera are combined with near-field array three-dimensional SAR system,and the data of three sources are fused.Using lidar’s ability to accurately locate and describe the shape and size of the target as well as the color and texture information of the optical image can effectively suppress the interference in the SAR image,accurately locate and diagnose the scattering characteristics of target,determine the target category,and improve the scattering diagnosis,detection and recognition ability of near-field array 3D SAR.To realize multi-source data fusion,the key is to solve the coordinate system alignment problem.Calibration and point cloud registration are two common methods.The former is coordinate system alignment of sensor,and the latter is coordinate system alignment of data.However,due to the large differences in imaging mechanism and imaging characteristics among SAR,lidar and camera,it is difficult for traditional calibration and point cloud registration methods to align coordinate system of three sources.Moreover,there have been no public reports on calibration and point cloud registration methods of near-field array three-dimensional SAR and other sensor data.Therefore,aiming at this problem,this thesis carried out an in-depth study,and proposed a multi-source fusion method based on calibration and a multi-source fusion method based on point cloud registration.The main work and innovation of this thesis are as follows:(1)A joint calibration method of near-field array three-dimensional SAR and lidar is proposed.Firstly,the problems faced in the calibration of near-field array threedimensional SAR and lidar are analyzed,especially the difficulty in finding calibration targets and selecting corresponding calibration points for the two systems.Then,the design,extraction and alignment of calibration targets are studied deeply,and a joint calibration method of near-field array three-dimensional SAR and lidar is proposed.Firstly,the target with high scattering and high reflectivity is designed as the calibration target.Then,to solve the problem of position deviation caused by high reflectivity,the MSAC plane fitting method which can overcome the interference of outliers is used to correct the position deviation.Finally,to solve the problem that the calibration target presents point cluster and it is difficult to select corresponding points,the processing method of calibration target point set matching is adopted,that is,CPD method is used to align the calibration target,and then the coordinate system of near-field array threedimensional SAR and lidar system is aligned.Finally,the alignment of three source coordinate systems is realized by combining the calibration method of lidar and camera.The experimental results show that the calibration error RMSE of near-field array threedimensional SAR and lidar is on the order of cm and this thesis realizes the multi-source data alignment in the experimental scene and verifies the effectiveness of the method.(2)A registration method of near-field array three-dimensional SAR point cloud and lidar point cloud is proposed.Firstly,the main problems of the calibration method are analyzed,such as the need to keep the relative position of the sensor in space unchanged,the design of specific calibration targets and the harsh acquisition conditions,which limited the flexibility of the method.For this reason,the registration method based on multi-source point cloud data is deeply studied.First,according to the characteristics of multi-source point cloud data,the corresponding point cloud preprocessing process is designed to extract the target point cloud from the collected data.Then,the registration method of near-field array three-dimensional SAR point cloud and lidar point cloud is proposed.In this method,a three-stage process of key point extraction,rough registration and fine registration is designed.Firstly,the key points are extracted from the target point cloud by an improved CED key point extraction method with the constraints of geometric structure and intensity.Then,the improved SAC-IA rough registration method with mixed constraints of trigonometric geometry and SHOT features is used to rough register the target point cloud,and the initial pose transformation result is obtained.Finally,based on the initial pose transformation,the improved ICP precision registration algorithm based on adaptive threshold is used to get the accurate pose transformation faster.For optical colored point clouds,PCA method is used to correct dimension distortion,and then the above method is used to register lidar point clouds and optical colored point clouds.Thus,the registration of near-field array three-dimensional SAR point cloud,lidar point cloud and optical colored point cloud is realized.Finally,the nearest neighbor search algorithm is used to remove the redundancy of SAR point cloud,and the multi-source point cloud fusion result is obtained.The processing results of multi-source data obtained by the experimental system verify the validity of the proposed registration method,and realize the registration of near-field array three-dimensional SAR point cloud with other source point cloud.