Study On Navigation Of Vision-Inertial Deep Integration For Proximity Operations With Space Target

On-orbit service is the hotspot of current research in aerospace field,and high precision measurement of non-cooperative targets,such as failed spacecraft,is the premise to realize the proximity operation of on-orbit service.According to the degree of non-coorperation,the non-coorperative target can be divided into two type,one is partial non-cooperative target which the geometric model is known,another is completely noncooperative target.Sponsored by the advanced programme of ‘xxx non-cooperative target visual measurement technology’,this dissertation aims at the problem of partial noncooperative target measurement and state determination in the far-range and near-range of on-orbit service by using visual camera and inertial navigation unit which are equipped on service spacecraft.This dissertation investigates the navigation method of the visual inertial integration,and deeply digs the modes of integration of visual camera and inertial measurement unit during each phases of proximity operation,and focuses on the in-depth study of the problems such as the restoration of blur star image,the attitude estimation of spacecraft by using the method of visual-inertial integration,the location determination of non-cooperative target in far-range,and the determination of the relative pose of noncooperative target in near-range.The main content is as follows:The restoration of blur star image caused by the movement of service spacecraft is a necessary operation of visual navigation of spacecraft.Blind restoration of image is an ill-posed problem and its calculated amount is large.A gyroscope aided method of blur star image restoration based on degenerate function is proposed.In this method,the star streak is assumed to be visible,the blur kernel for image restoration is calculated by using the angular velocity of gyroscope,and the matrix model of the centerline of star streak is constructed in the original image.On this basis,the centerline of star streak is taken as the observation and the sparsity of blur kernel is taken as the constraint.The least squares optimization model for blur kernel correction is constructed,and the pre-conditional conjugate gradient internal point method is used to achieve the optimal solution for global convergence.Finally,the simulations are carried out.The results show that the accuracy of blur kernel is significantly improved after correction.Moreover,the position accuracy and recognition rate are improved after restoration.The accuracy of attitude of service spacecraft will affect the accuracy of the measurement of non-cooperative targets,so the output information of visual camera and gyroscope are needed to be integrated deeply,and the measurement delay of visual camera should be considered.Based on the factor graph theory,a method of visual inertial integration estimation for the attitude of service spacecraft is proposed.In this method,the attitude factor model of visual camera with measurement delay and the gyro factor model with bias are constructed.The optimization model of visual-inertial integration is established based on the joint probability density,and the solution is optimized by incremental smoothing based on sliding window.In order to improve the calculation speed,the measurement of camera is divided into keyframe and non-keyframe.In nonkeyframes,the factor of visual camera is constructed by the star point coordinates in image,which is used to modify the keyframe optimization model.As the complex star matching is avoided,the efficiency of the attitude estimation is greatly improved.The mathematical simulation results show that the proposed method can deal the measurement delay of visual camera which causes deviation to the attitude estimation result,and highprecision attitude data with high frequency can be output.To detect and locate the target spacecraft when the distance of two spacecrafts is far,the inertial measurement information of the service spacecraft is introduced,and a noncooperative small target detection method based on the feature of the motion trajectory is proposed.After the rapid segmentation of light point,the motion trajectory of stars in sequential star images by using the angular velocity of inertial measurement unit is calculated,and the star motion in image caused by the movement of service spacecraft is eliminated.After that,the non-cooperative small targets are detected.According to the model of relative motion,the target tracking based on the prediction of location aided by inertial measurement unit is used to solve the problem of target false recognition caused by the new star which enters the field of view for the first time.Finally,the factor graph model for the relative motion state estimation of the two spacecraft is constructed,and the relative motion trajectory and relative velocity are estimated.The simulation results of real star images show that the efficiency of the proposed method is higher than that of the images difference method and the method of location estimation when the priori information of the target position is unknown.The feasibility of relative motion trajectory estimation method is verified by mathematic simulation.In order to determine the relative pose of partial non-cooperative target whose geometric model is known in near-range,by using the circle feature at docking ring or nozzle on spacecraft,an elliptic feature extraction method based on texture boundary feature is proposed,which under an assumption that the grayscale distribution of one component is the same.Based on the condition probability model of elliptic texture boundary,the probability expectation of the elliptic feature pixel point on the texture boundary is calculated,and the ellipse function is fitted.In order to improve the accuracy of elliptic extraction,inertial navigation information is introduced to predict the position of ellipse.Combining the information of elliptic feature and inertial measurement,the normal vector and center position of spatial circle feature are given in analytic form.The simulation results show that the method can accurately extract elliptic feature when the texture of spacecraft surface is complex,and it is robustness to noise.The performance of the proposed method is better than the method based on edge feature and gradient feature.The simulation verification of the relative pose is carried out,and the influence factors of the relative pose error are analyzed.