Binocular Vision Pose Estimation Algorithm Research

This paper researches binocular vision algorithm of the spacecraft relativeposition and orientation estimation, which is applied on Space Rendezvous andDocking of spacecraft， and which is based on computer vision measurementtechnology. The achievement of Space Rendezvous and Docking technology has greatsignificance for aerospace engineering. It is a key technology for China’s space project.The binocular vision algorithm is still in its infancy. Therefore, the subject has stillmuch work to do. This paper is in this context.The binocular vision measurement technique is an important research direction incomputer vision. It includes image acquisition, camera calibration, feature extraction,stereo matching,3D reconstruction and motion estimation in six parts. This studyfocuses mainly on the back of four steps.Firstly, this paper introduces the mathematical models and geometric relationshipsof the binocular vision system. This is theoretical basis of binocular vision algorithm,which paves the way for the subsequent algorithm. According to the shapecharacteristics of the spacecraft, we take the corner of image as feature information.Firstly, preprocess image by smoothing filter to improve signal to noise ratio of theimage. Then analyze and discuss four kinds of corner detection algorithm principle andimplementation steps, and make reasonable improvement. The stereo matching isvery difficult process in binocular vision. This paper designs algorithm based on thecharacteristics of the image point according to the application requirements of motionestimation. And then take the epipolar constraint strategy to exclude mismatchingpoints in gray correlation matching，which approve the veracity observably. The3Dreconstruction process is based on the parallax principle of binocular vision. The motion estimation is also pose estimation. Using criterion of the sum of squared errorsminimization infers two estimation algorithms: the unit quaternion method andsingular value decomposition method. The simulating results show that these twoalgorithms can achieve high estimation accuracy, where the relative error range oftranslation vector is less than2%and the error range of the rotation angle is less than1°. It can meet the application requirements of spacecraft rendezvous and docking.