Research On Pose Measurement Of Space Non Cooperative Target Based On Binocular Vision

With the further exploration and development of human beings in outer space,all countries put forward higher requirements for space technology,such as satellite spacecraft rendezvous and docking,the fault between the on orbit capture and maintenance and space debris removal.They have become an important problem to be solved for the development of the aerospace industry,and grasping rendezvous and docking with non cooperative target is the key technology to solve these problems.In order to achieve this goal,it is necessary to solve the relative position measurement problem between non cooperative targets.At present,the measurement methods for non cooperative targets are mainly based on monocular or binocular pose measurement method,position and pose measurement method based on scanning laser radar,pose measurement method based on scanless 3D imaging and pose measurement method based on multi-sensor fusion and so on.The binocular system composed of two identical cameras can simulate the human eye's imaging process,and the system uses triangulation method to obtain the three-dimensional information of the target relatively easily.Therefore,binocular system also often takes the pose measurement of non cooperative targets.The binocular system involves the binocular stereo vision technology,and the technology is a branch of computer vision technology,its principle is to shoot the target object two or more images from different angles,and calculate the position deviation of image pixels using the geometric imaging principle(i.e.visual difference)3D information to obtain the target.To restore the real 3D scene graph corresponding to the target.Therefore,this subject uses binocular vision to measure the non cooperative target of space,and studies the key technology of the subject in this process.The main work of this article is as follows:First,this paper introduces the coordinate system and the camera imaging model of vision measurement involved,and then influence the configuration parameters of binocular stereo system for measuring object attitude measurement accuracy is analyzed by theory,and get some conclusions,these conclusions based on binocular vision for camera configuration of non cooperative target attitude measurement has theoretical significance and reference value.Then,the influence of camera calibration error and stereo matching error on pose measurement is analyzed by numerical simulation.Secondly,the method of camera calibration is studied in this paper.At present the Zhang's calibration method has some shortcomings,the main point is the nonlinear iterative optimization of many parameters,coupling the distortion center and the distortion coefficient will make iterative optimization results unstable,on the one hand,the correct calculation of the distortion center and distortion coefficient may be adjusted to the internal parameters of the camera;on the other hand,the computational efficiency will be affected.In this paper,a camera calibration method is proposed to decouple the distortion coefficient and internal parameters,and the distortion center is also accurately calculated by this method.By introducing the division distortion model,we first solve the estimation of the distortion coefficient and homography matrix(nonlinear problem)through the linear least square method.Then the inner parameters of the solution are used as the initial value for nonlinear iterative optimization,and the higher accuracy is obtained.The simulation experiment and the actual experiment show that the method in this paper has the advantages of simple,flexible and high precision.Then,this paper studies the existing stereo matching method.The stereo matching method with epipolar line constraint reduces the search space from two-dimensional plane to one-dimensional line,which not only improves the search efficiency,but also improves the accuracy of stereo matching.The polar line constraints can be represented by a basic matrix,which describes the polar geometry of the two images taken from different perspectives.In order to improve the precision of stereo matching,a fast and robust basic matrix estimation method is proposed in this paper.The method abandoned the traditional robust estimation method to separate the field values and then calculate the basic matrix.Instead,it eliminates the outliers in the basic matrix estimation process.The potential error corresponding points are eliminated by iteration,thus the stable estimation of the basic matrix is realized.Then,in every iteration process,we use a criterion of polar geometric error to identify outliers,so that we can get the estimation results of fundamental matrix at the same time.The iterative process of this method converges quickly.Even if there are a large number of matching field values,the final results will soon be stable.The simulation and the actual experimental results show that the proposed algorithm has a great improvement in computing efficiency while guaranteeing similar estimation accuracy.Besides,the present position and attitude measurement methods are summarized in this paper.In view of the shortcomings of traditional position and attitude measurement methods,some improvements have been put forward.First of all,there are some problems in the current methods of non iterative pose estimation,or the accuracy of pose estimation is relatively low when the number of target reference points is small,or the performance of estimation is seriously degraded when the target reference point is quasi singular or planar.In this paper,by using the rotation matrix parametric representation,and propose pose estimation into solving the objective function minimization problem can be applied to the n>=3 arbitrary points and different configurations.It has strong stability,and the estimation accuracy is better than the existing method.Then,the existing iterative pose estimation method has some problems.Almost no consideration is given to the constraints of the target reference point itself.In addition,when the target reference point is quasi singular or planar,the estimation performance is seriously degraded and iterative optimization may not converge or converge to a local optimal value.In this paper,a pose measurement method based on geometric constraints is proposed.This method takes full account of the constraints between target reference points,and applies to different configurations of reference points.It has strong stability and the accuracy of estimation is better than the existing iterative pose estimation methods.Finally,this paper sets up a platform for the binocular vision measurement system.The proposed method is used to measure the space non cooperative target model,and the location and attitude angle of the space non cooperative target model relative to the measuring coordinate system is obtained.