Research On Pose Estimation For Space Non-cooperative Targets Based On Monocular Vision

With the progress in space technology,the space operation technology,which is widely used in on-orbit spacecrafts maintenance,as well as in space debris removal,has gained rapid development.The targets for space operation are usually non-cooperative spacecrafts which do not provide any effective information.As an critical step to dock or capture the uncooperative space objects,achieving the accurate position and attitude measurement for non-cooperative targets play an important role in space operations.This dissertation focuses on the demand of future space operation tasks,and studies the technologies for the pose estimation of uncooperative space objects with the background of vision measurement application in space operations.The main work is as follows:After the deep research of the technology of camera calibration,an efficient non-iterative calibration method is proposed in this dissertation based on the radial distortion division model.By using this method,the centre-of-distortion is accurately estimated.Moreover,it decouples the estimation of the distortion coefficients and intrinsic parameters,yielding more stable and reliable solution.In addition,the effect of lens distortion has been taken into account when solving the homography matrix.Therefore,the proposed method can obtain high accuracy calibration result without the nonlinear iterative optimization procedure.Compared with Zhang's method,it can improve the computational efficiency by about 10 times under the premise of ensuring the same calibration accuracy,thus achieves fast and accurate camera calibration.Compared with the cooperative target,the non-cooperative target lacks the prior knowledge of the 3D structure of the target,and does not have the artificial mark which is easy to be reliably extracted and identified.Therefore,this dissertation proposes to reconstruct the 3D feature points of the target from the sequence images.Through the establishment of the 3D feature points of the target,the non-cooperative target is transformed into the cooperative target.Considering that there are large scale and rotation changes in the actual measurement target,the SIFT local feature are used to reconstruct the 3D feature of the target,and the 3D structure model for describing the target is established steadily and effectively.For the computation of two-view epipolar geometry,a fast method for fundamental matrix estimation is proposed.It is inherently robust and does not require the preprocessing steps based on hypothesis testing to remove outliers.Moreover,the outlier rejection scheme is integrated within the fundamental matrix estimation pipeline with a negligible expense,resulting in an approach with high computational efficiency.Then,this dissertation researches on 3D reconstruction technologies from image sequences based on SFM.And the reconstruction and optimization of the 3D structure of the target feature is achieved by combining the two-view feature point reconstruction and the bundle adjustment method.The experimental results show that this reconstruction method can effectively recover and describe the 3D feature points of the object,which solves the problem of lack of effective cooperative information on non-cooperative targets.In order to realize the measurement of the pose of the target,this dissertation makes a deep research on the pose estimation algorithms based on the feature points,and proposes an improved non-iterative method.By introducing the measurement error model,the generalized pose estimation problem is formulated as a minimization of a least squares cost function.The proposed method is applicable to a PnP problem with any point number n(n ? 3).Moreover,it has high stability that can effectively deal with different point configurations robustly.The method computes all pose solutions analytically as the minima of a nonlinear least-squares cost function,without the need for initialization.Compared with the existing pose estimation methods,it shows significant advantage in the estimation accuracy,especially for small point sets.On the basis of the above research results,this dissertation presents a method of pose measurement for non cooperative targets based on SIFT feature matching.In this method,the 2D-3D correspondences between the feature points in the image and the 3D points in the target model database is established by matching the SIFT feature descriptors.Then the robust pose estimation algorithm is used to solve the corresponding pose parameters of non-cooperative targets from the 2D-3D correspondences.Finally,the influences of various factors on the final pose measurement result are analyzed by numerical simulation,and the experimental platform is set up for the pose measurement experiment,verified the feasibility of the proposed pose measurement scheme for non-cooperative targets.Aiming at the requirement of the attitude measurement for uncooperative space objects,this dissertation studies the related methods and the ground verification of the key technologies.In this dissertation,several non-space objects instead of the real space targets are used to verify the feasibility of the proposed scheme.And it lays the theorical and technical foundation for the future attitude measurement of space non-cooperative targets.