Research On Relative Visual Navigation Method Of Spacecraft Based On Natural Features

On-orbit service of spacecraft is a hot spot and frontier direction in space field of home and abroad.High precision autonomous relative navigation is the precondition of approach and control for on-orbit service spacecraft.Monocular vision has become the preferred sensor for on-orbit spacecraft relative navigation with advantanges of small volume,low power consumption and abundant information.So,relative navigation method and technology based on monocular need to be explored and the related research for spacecraft such as failed spacecraft which is without cooperative markers is imminent.Sponsored by the the project of ‘xxx on-orbit service spacecraft visual measurement technology' of a ministry,this thesis researchs the active feature selection,robust relative pose determined and vision/IMU combined navigation taking on-orbit service spacecraft as the research background and considering problems of feature redundancy,the existence of outliers,low output frequency and discontinuous measurement using monocular for navigation based on natural features of the target spacecraft.The innovative research results obtained are as following:Natural features extracted cantain a large number of redundant features and using all the features for navigation will result in large amount of calculation.Thus,a strategy to select effective feature subset which decide the navigation performance is proposed.Two methods for selecting natural features actively are given based on convex hull and minimum redundancy respectively.Based on the definition of the dilution of precision for visual system related to the number and distribution of features and the analysis of the influence on the navigation precision,a two-stage solution algorithm for solving the convex hull of natural features with the largest area is designed.Natural features selected by the algorithm satisfy the requirement of geometric distribution precision.Then,on the basis of defining the dispersion of two points by using the angle between projection vectors of two image points,and constructing redundancies of natural features,a active features selection method by a iterative process based on the minimum redundancy is proposed,which can effectively realize to select the given number of natural features.Mathematical simulations show that both methods can select features with good geometric distribution.When the number of features is reduced to less than 1/4 of the original number,the solution accuracy of selected features is not significantly reduced.When natural features are used to determine relative pose of spacecraft,the result may produce gross error and even be invalid since the existence of outliers.To slove this problem,a robust method to determine relative pose of spacecraft based on kernel space is proposed.The method constructs the constraint equation in the tangent space of all feature points and defines the geometric error of feature points using the kernel space of constraint equation.Then,it builds the threshold which is used to judge outliers and elimitates outliers by an iterative method.The kernel space is then estimated robustly.Considering the constraint of the rotation matrix,a nonlinear optimization method of relative pose determination based on orthogonal iteration and an analytical solution of initial solution are given.Mathematical simulations show that the proposed method can effectively eliminate outliers and determine the relative pose with high accuracy.The proposed method improves the robustness of the relative pose determined with natural features and the time taken is less affected by the proportion of outliers.The effectiveness of this method is further verified using real images.The frequency of monocular is slow and the measurement of the visual sensor is discontinuity caused by external interference such as specific illumination.The relative navigation method with monocular camera and IMU is explored by taking advantage of IMU's high output frequency and small environmental impact.The installation pose between the two sensors need to be calibratied.The off-line calibration method of camera and IMU installation pose requires additional equipment and the calibration process is complex.Thus,a vision /IMU combinated method for relative navigation and on-line self-calibration of spacecraft is proposed to solve the problem.The system observation model is established considering the relationship between camera/IMU installation pose and the camera measurement.A deviation correction filtering algorithm is designed to suppress linearization error by using the extended state variables including installation pose.Then,the system observability is analyzed by using nonlinear observability theory.Finally,the effectiveness and correctness of the proposed vision/IMU relative navigation and self-calibration method are verified by mathematical simulations.