Research On Motion Estimation For Relative Navigation Based On Computer Vision

The study of computer vision navigation is concentrated on vehicle navigation atpresent. Vision navigation is applied to the spacecraft autonomous landing which hasbecome the research direction of many scholars. This paper studies the estimationproblem of motion parameters in vision navigation around the autonomous landingprocess of spacecraft and enriches the application of computer vision in spacecraftlanding navigation. As follows, the achievements and main research contents aresummed up:1) The fundamental matrix is the basic tool of analysis of the two images whichare two uncalibration images from the same scene. The first step of visual applicationsuch as analysis of motion and camera calibration is the estimation of fundamentalmatrix. Paper gives an improved algorithm for estimation of fundamental matrix basedon the problem of analysis of motion in visual navigation. LM (Levenberg-Marquard)method initialized by optimally corrected ML obtains the solution of fundamentalmatrix. And some simulation comparison experiments for different algorithm havebeen carried out by selection simulated images and real images. The results ofexperiments indicate that our improved algorithm has the higher efficiency andaccuracy.2) The analysis of motion for objective needs the detailed description of the errorof measurement in the visual navigation system. The sensor of computer vision hasthree different types of errors nature in the process of imaging. The first is thenonlinear anamorphose of optical system, which in the process of camera calibrationcan be corrected in advance; The second is the noise of image, which can affect theveracity of feature detection. But its role is usually only limited to sub pixel level. The last is the characteristic pixel position error of the digital image due to the spatialquantitative effect of sensor. The key points of our research is the image noise. In thispaper, the essential matrix is decomposed by factorization algorithm then to getexternal parameters. And the external parameters are got by the Singular ValueDecomposition theory. Then the error propagation of motion estimation is analyzedbased on covariance propagation theory. Simulation experiments of derived resultshave been carried out by using simulated image. Experiments show that the derivedresults are correct.3) The position and posture of spacecraft are solved by the epipolar geometricconstraint in the visual navigation system, however the precision and stability of thismethod is not good. This paper introduces the coordinate system in the process ofanalysis of motion and establishes the filtering model of visual navigation system.Then the central difference kalman filter is applied to the visual navigation system.Some simulation comparison experiments have been implemented between thisalgorithm and the epipolar geometric constraint. And this algorithm has the higheraccuracy and stability and can estimate the spacecraft’s relative position and attitudeinformation well.