| Benefitting from computer and image processing technology, the vision measurement technology plays a most important role in the fields of aerospace, industrial production and medicine. The binocular vision has a great advantage of accuracy over monocular vision. The paper starting from practical application, introduced the problem of estimating the relative positions and attitudes of active and passive targets.Firstly, a matching scheme based on checkboard pattern was designd to fufil the task of feature matching in the study of estimating the relative positions and attitudes of active target of binocular vision,which extracting the black and white rectangle vertex as Harris corners, after which a series of technology was used such as line detection, relative position arrangement and subpixel computing.Proposed a method of estimating the relative positions and attitudes of active target, which extended the traditional PnP method, that can estimate relative positions and attitudes from the information of feature points of left and right camera without feature matching. Study the hand-eye calibration, improved the stability against noise from a series of action information of calibration platform as the calibration data,using the least-squres method based on RANSAC, which also improved the characteristics of rotation matrix with SVD decomposition in computing the relative pose between hand and eye. Then, did experiment on the calibration platform, after which analysed error from100data, which proved the proposed method in accuracy, real-time and stabilityAt last, extracted the feature of real passive target in space rendezvous, analysed how to choose the feature of passive targets, proposed a scheme which can detect ellipse fast, and used a series of image processing technology such as contour extraction, edge detection and ellipse fit. Extracted two ellipse feature of target stably and accurately. Introduced a method of pose estimation based on this planar ellipse feature. |
PDF Full Text Request