Research On The Pose Estimation Method Of Space Targets Based On Ground-based Optical Telescopes

With the development of the technology of large aperture optical telescopes and adaptive optics system,the resolution of space target observation images can reach the diffraction limit,which provides effective data for space target detection and recognition.Important target characteristics such as size,pose,shape,orbit parameters and so on can be obtained by analyzing high resolution spatial target images.Among them,the pose estimation result of the target can be widely used in satellite classification and recognition,mission state evaluation,detector direction determination and abnormal behavior detection,etc.The research of pose estimation method of the space target is an important research topic.Based on the comprehensive consideration of the existing pose estimation methods and characteristics of optical telescope observation data,a pose estimation method by real-time model pose matching is proposed based on practical engineering applications.Based on the comprehensive consideration of the existing pose estimation methods and characteristics of optical telescope observation data,a pose estimation method by real-time model pose matching is proposed for practical engineering applications.This pose estimation method is mainly used to analyze the observed targets with known the 3D model and orbital elements of the target.First,the observation image is restored to improve the accuracy of pose estimation.Then the observation ideal image is obtained by the real-time simulation of the observation process of the telescope.Finally,the pose of the target model is adjusted by the pose search strategy until the simulated observation image and the actual observation image are most similar.The process of the pose estimation mainly involves three key technologies: the fast image restoration technology,the space target in-orbit simulation observation technology and the real-time model pose matching technology,which are deeply studied in this paper.(1)Fast image restoration techniqueBased on the multi-frame blind deconvolution(MFBD)algorithm,the cost function is parallelized to improve the parallelization of the algorithm.The parallel running method of GPU cluster is designed,and the parallel improved MFBD program is deployed on the GPU computing cluster to realize the fast restoration of the observed images.(2)The technique of simulation observation of the space target in-orbitBased on the technology of visual simulation,a simulation observation system of the space target in-orbit is designed and developed.According to the satellite model,satellite ephemeris data and optical system parameters,the simulation ideal image in accordance with the actual observation effect is obtained by simulating the observation process of the satellite based on ground-based optical telescope.The target model pose can be adjusted in real time to obtain the simulated observation images of different target pose.(3)Real-time model pose matching techniqueA relative pose estimation method of the space target based on the real-time model pose matching is proposed.The maximum similarity measurement between the simulated observation image and the actual observation image is solved by the gradient search method.When the similarity between the simulated image and the actual observation image is optimal,the pose of the target model can be regarded as the pose of the real target.The simulation results show that,under ideal conditions,the model pose can converge to the real value when the input initial pose value is in the effective range.