Research On Real-time Measurement Technology Of Non-cooperative Target Position And Pose Based On Stereo Vision

With the diversified development of space missions,the human demand for spacecraft in-orbit service is becoming more and more urgent,among which non-cooperative target position measurement is one of the key technologies in the fields of fault repair and in-orbit capture,which has important research value and significance.Among the non-contact measurement methods,stereo vision system is able to acquire 3D stereo information without special light source,which has been a popular research problem in vision measurement.However,the considerable amount of image data computation,special scene environment,limited payload and other factors as well as the requirements for system real-time,robustness and integration have posed great challenges for the practical application of stereo vision systems in space missions.In this paper,the problem of non-cooperative target attitude measurement in on-orbit service near operations is investigated in the context of spacecraft on-orbit service.Firstly,a stereo vision system model for non-cooperative target attitude measurement is developed in this paper.The vision model is deeply analyzed by the conversion problem between coordinate systems and the basic principle of camera calibration is investigated.Subsequently,the binocular camera calibration and correction experiments are conducted to reduce the impact of lens distortion and other factors on the subsequent visual image processing and pose measurement.Secondly,a time-domain correlation-based binocular stereo matching algorithm and its hardware computing architecture are proposed to address the problems of accuracy,robustness and real-time performance of stereo matching algorithm in 3D reconstruction.By improving the traditional binocular semi-global matching algorithm,the calculation results of the previous frame are introduced into the current frame calculation to match from the whole time series,and the FPGA(Field-Programmable Gate Array)hardware computing architecture of the algorithm in this paper is designed,and then the accuracy,robustness and real-time performance of the algorithm and architecture in this paper are verified by data set experiments and real scene experiments.The accuracy,robustness and real-time performance of the algorithm and the architecture are then verified by data set experiments and real scenario experiments.Then,a point cloud feature-based pose measurement algorithm is proposed for the noncooperative target point cloud model pose measurement problem.The target point cloud model is generated and preprocessed based on the depth images obtained from binocular vision 3D reconstruction,and the geometric topological relationships are recovered by surface reconstruction using triangular dissection,and then the target principal axes are estimated by extracting the point cloud center of mass and surface normal features and optimized with the traceless Kalman filter.Finally,in order to further improve the integration of the measurement system,a MPSoC(Multiprocessor Systems-on-Chip)based real-time positional measurement system is designed.The on-chip FPGA accelerates the stereo matching calculation to achieve 3D reconstruction,and the on-chip ARM processor performs point cloud data processing and target attitude measurement.The overall experimental scheme is designed,and the effectiveness of the measurement system is verified through the posture measurement experiments.