One of the important task for a computer vision system is to recover the structure of objects in the scene from their images. In practice, since many objects( especially man-made objects) have conics on their surface or have quadric surface, we propose some 3D reconstruction algorithms for conics and quadric surface using conics as elementary corresponding feature. By combining these methods with the ones, such as pose determination, reconstruction methods for polyhedral, we believe that we can build a vision system for the industrial application. The thesis is concentrated on the subject " Research of Computing Theory for Computer Vision". In the thesis, we not only propose some simple but workable reconstruction algorithms, for example, the stereo algorithm of conics in Chapter 3 has closed-form solutions, the stereo algorithm of quadric surface only need solve a system of three quadratic equations, in the same time, we also bring some new ideas in the field, such as hypothesis-verification, active vision, using redundant information etc., to improve the robustness of our algorithms. All of the above measures in some extent make our algorithms more applicable in practice. All the algorithms in the thesis are tested by numerous experiments on real data.
修改评论