Non-Cooperative Target Pose Estimation Based On Stereo Vision

With the deep exploration of space environment,the types of space activities and tasks have been developing in a diversified direction.Among them,space debris removal,space satellite rendezvous and docking,on-orbit maintenance of spacecrafts and other activities have become important issues that need to be solved in the development of space technology.In these spatial activities,the acquisition of non-cooperative target pose information is the key to accomplishing the task,and stereoscopic pose measurement technology as an effective method for non-cooperative target pose estimation has received wide attention from all countries.In view of the spatial non-cooperative target relative pose estimation task requirements,this paper conducts an in-depth study on the spatial non-cooperative target pose estimation technology based on binocular vision.The main research work includes the following aspects:In the aspect of camera imaging model and camera calibration,the related coordinate systems conversion and camera imaging model are introduced firstly.Then the binocular system stereo calibration method is analyzed based on the camera imaging model.Finally,the binocular system stereo calibration experiment is carried out using the Matlab Calibration Toolbox.The results of calibration provide support for further solution of target point cloud data.In the aspect of stereo matching and point cloud data solving,firstly,there are a lot of mismatching problems for traditional feature extraction and matching algorithms.A method for further eliminating mismatched points by using polar line constraints is proposed,which improves the accuracy of stereo matching.The mathematical relationship model between image feature points and spatial point coordinates is then constructed,and the fast solution of the target point space coordinates is realized.Three accuracy evaluation methods are proposed for the acquired point cloud.The results show that the projection error is less than 0.1 pixel averagely,the point cloud size error is less than 3%,and the target distance error is less than 1.5%.In the aspect of relative pose estimation,the classic two-step method is used to achieve high-precision registration of point clouds.Firstly,the Sampling Consistency Initial Registration Algorithm(SAC-IA)is applied to realize the initial registration of the point cloud,which provides a better initial position for further accurate registration.Then,the point clouds after initial registration are accurately registered using the Iterative Closest Point Algorithm(ICP)to obtain the relative pose information.In order to verify the ability of the target pose measurement algorithm,3DsMax simulation software is used to simulate the flying process of binocular stereo system,and a series of non-cooperative target pose measurement simulation experiments are carried out based on the image sequence.The experimental results show that the attitude measurement error is kept within ±0.4°,and the position measurement error is kept within ±18cm.mearsuring at a distance of 30 m The algorithm in this paper can accurately estimates the pose of non-cooperative targets.