Research On 3D Modeling And Intelligent Identification Of On-orbit Service Spacecraft Based On Point Cloud

On-orbit service technology has become a research hotspot all over the world,which is aimed at maintaining the normal operation of space orbiting satellites and reducing the losses caused by collision with failed satellites.Aimed at the problem of capturing and repairing spacecraft components in the process of on-orbit service,this paper performs point cloud modeling on the target spacecraft to identify the components and types of the spacecraft.Firstly,depth image is converted into point cloud by spatial coordinate transformation,which is based on the working principle of the depth camera.Straight-through filtering,outlier filtering,and grid center-of-gravity reduction algorithm are used to denoise and reduce target point cloud,which can reduce the storage space of 3D model of the target spacecraft and provide high-quality point cloud data for later 3D modeling and intelligent recognition.Secondly,a multi-view point cloud stitching algorithm is designed for 3D modeling of target spacecraft.SAC-IA algorithm based on FPFH is used for initial registration.ICP algorithm based on the curvature feature is designed for precise registration.Based on the registered point cloud data,an improved point cloud stitching strategy is proposed.Multi-viewpoint cloud stitching of the registered point cloud data is performed by averaging the point coordinates of overlapping regions,which improves the accuracy of point cloud stitching,which lays foundation for later intelligent identification operations.Thirdly,for the intelligent identification of target spacecraft,the technology of spacecraft component identification and type identification is studied separately.studied separately.A normal vector-based RANSAC algorithm is proposed to identify different parts of the spacecraft.Experimental results show that the improved algorithm is of higher recognition accuracy and faster recognition speed than traditional RANSAC algorithm.Point Net deep learning neural network model is used to train and test self-built Model Net40＿Aero dataset,it can identify target types such as body-mounted satellites,expanded satellites and space debris.Experimental results show that the algorithm can recognize the types of different space targets.Finally,in order to verify the effectiveness of the proposed algorithm in practical applications,a point cloud modeling and intelligent identification verification platform for the target spacecraft is designed.The motorized rotating table is used to drive the spacecraft to rotate,which simulates the attitude of the spacecraft,and Kinect V2 is used to perform multi-view data acquisition on the spacecraft.The demonstration experimental platform software is developed in the Visual Studio performs point cloud data acquisition,modeling,and identification of the target spacecraft.The software effectively validates the aforementioned algorithms and provides a convenient operating interface to assist space on-orbit service tasks.