Research On Point Cloud Feature Recognition And Pose Measurement For Space Non-cooperative Targets

With the increasing growth of human space exploration activities,the total number of service or failures spacecraft in orbit is increasing.In order to meet the needs of future space exploration,it is urgent to carry out on-orbit service missions such as fueling of long-term service spacecraft,maintenance of faulty spacecraft,and clearance of failed spacecraft.The service spacecraft needs to obtain the accurate pose information of the target spacecraft relative to itself when performing the aforementioned on-orbit service mission,and most of the target spacecraft are space non-cooperative targets that cannot actively provide pose information.Therefore,using the observation information of the measurement sensor to identify the space non-cooperative target and measure its relative pose information has become a key prerequisite for the service spacecraft to perform the on-orbit service mission.Observation information can be divided into two-dimensional images and three-dimensional point clouds according to the expression dimension.Compared with the former,the three-dimensional point cloud can express depth information,provide more direct and more stereoscopic target information,The acquisition of point cloud,which has low noise,high resolution and high frame rate,becomes more and more convenient with the development of stereoscopic perception technology.Because the original three-dimensional point cloud in the space scene contains interference data such as target unrelated points and equipment noise points,it is necessary to identify and extract the target point cloud data in the scene according to the characteristics of the spatial non-cooperative target point cloud.The position and attitude information of the target spacecraft relative to the serving spacecraft is measured on the basis of the point cloud of the target spacecraft,which provides key support for the successful execution of the on-orbit service mission.Therefore,this paper takes space non-cooperative target spacecraft as the research object,aims at the pose measurement problem in the on-orbit service mission,and conducts research on the construction of 3D point cloud,point cloud feature recognition and pose measurement method.The specific research contents are as follows:Firstly,based on the depth camera,research on the construction of 3D point cloud of non-cooperative target in space is carried out.Establish a depth camera measurement model based on the principles of binocular vision,structured light,and time of flight;analyze and compare the measurement range and accuracy of the three measurement models,and select the appropriate depth camera measurement model in conjunction with the on-orbit service mission scene to collect and construct the scene Point cloud;build a spatial index structure of point cloud based on KD tree,improve the proximity search efficiency of point cloud discrete data points;perform preprocessing such as point cloud filtering,down-sampling and clustering segmentation on the original scene point cloud data to reduce point cloud data size and provide a basis for subsequent research.Secondly,based on the feature matching method,research on the feature recognition of point cloud of non-cooperative target is carried out.In order to obtain accurate recognition results of spatial non-cooperative targets,a spatial non-cooperative target recognition framework is constructed based on point cloud feature matching methods;point cloud VFH feature descriptors are analyzed and constructed,and point cloud VFH feature sets are constructed based on spatial non-cooperative target CAD models,providing a data foundation of feature matching;in terms of the problem of slow recognition speed caused by large-scale feature collection,a rotationally symmetric feature detection method is introduced in the spatial non-cooperative target recognition framework,which analyzes the rotationally symmetric feature of the target and removes redundant viewpoints in the process of constructing the VFH feature set accordingly to reduce the size of the feature set.The improved non-cooperative target recognition method can achieve accurate recognition of non-cooperative targets,the recognition speed is increased by 63.63%,and the efficiency of target recognition is greatly improved.Then,based on the point cloud registration method,the spatial non-cooperative target pose measurement research is carried out.Build a non-cooperative target pose measurement model to describe the spatial coordinates transformation relationship;uses the SAC method for the initial registration of the non-cooperative target scene point cloud and reference point cloud to reduce the initial pose error,aims at the problem of long computation time,improve the SAC method through edge sampling point selection strategy and improve the registration efficiency up to 87.18%;by using ICP method for accurate registration,and construct a point cloud registration overlap rate error evaluation index for the problem of the registration result falling into local optimization easily,evaluating the registration effect from multiple angles to improve the reliability of the non-cooperative target pose information.Finally,an experimental verification research is carried out on the point cloud feature recognition method and pose measurement method proposed in this paper.Based on the scale model of the target spacecraft and the Kinect v2 depth camera,an experimental platform was built to carry out space non-cooperative target point cloud feature recognition and pose measurement experiments;comparative analysis of the experimental data to verify the feasibility and effectiveness of proposed recognition and posture measurement methods.