| Along with human exploration of outer space,the environment of space is becoming increasingly complex with the growing number of spacecrafts.In recent years,the security of space assets and national information are facing unprecedented challenges.At present,the traditional location determination of space targets based on orbit information cannot meet the practical needs of space activities,and the demand of accurate estimation of shape and pose of space targets is increasingly strong.Currently,the images from optical imaging equipment and ground-based radar can supply the pose and shape estimation of space targets.However,the accuracy of pose and shape estimation is limited by the observation angle,the degradation of observed images and the weak texture of space object material.In order to solve the above problems,this paper aims to improve the accuracy of the shape and pose estimation with the support of prior shape information and multi-source observation.The main work and innovation of this paper are as follows:Aiming at the pose estimation when the parameters of target model are known,the method of pose estimation of space targets based on fixed parameter model is proposed.The basic structure of space targets is considered as a combination of parameterized geometric primitives in this paper.The traditional artificial features show poor robustness in the special observation scene of space.This paper devotes to establishing correspondence between the 2D image and 3D model by deep learning technology,and on this basis the accurate pose is obtained.The proposed method performs better in the efficiency and accuracy compared with the existing methods.Besides,in order to eliminate the fuzzy in keypoint detection caused by weak texture material and symmetrical structure of space targets,the candidate space is established in the parameter estimation process,which improves the accuracy of pose estimation under high dynamic illumination and fuzzy degradation.Aiming at the pose estimation when the partial parameters of target model are unknown,the method of pose estimation based on holistic constraint is proposed.Due to the fact that the shape of space targets may change in orbit,the deformable parameter model is established to describe the shape of the target,and an iterative optimization method is used to estimate the parameters of shape and pose based on the extraction of keypoints.Then considering the lack of local information under extreme illumination conditions,a method of parameter estimation combining local keypoints and global center points is proposed to improve the accuracy of pose estimation when the shape of target changes.Aiming at the pose estimation when the parameters of target model are unknown,the method of pose and shape estimation based on hierarchical model is proposed.For unknown space targets,it is difficult to estimate the shape of target only by images under special observation conditions in space.To this end,the hierarchical model is used to describe the structure regularity,and then the estimation of shape and pose of space targets is conducted based on the hierarchical model.The proposed method improves the accuracy and completeness of shape estimation of unknown space targets under special space observation conditions.Aiming at the pose and shape estimation when in the case of multi-source data collaborative processing,the method of shape and pose estimation of unknown space targets from the optical and ISAR images is proposed.In response to challenges brought by the strong heterogeneity between optical and ISAR images to collaborative processing,the framework of shape and pose estimation of the space targets combining optical images and ISAR images is proposed.In the proposed method,the parameters of shape and pose are obtained through the iterative process of model parameters between the two kinds of observation images,and the estimation accuracy of shape and pose is improved combining optical and ISAR images.Finally,the task flow and instance of shape and pose estimation of the space targets is given based on the researches in this paper. |
