多视几何continue.ppt

如何从F中恢复P,P’? 如何从F中恢复P,P’? 如何从F中恢复P,P’? canonic camera matrices: Rank=2, 给定一基本矩阵 F，构造投影矩阵对 2. 两视几何 有了投影矩阵和图像点就可以通过三角化(Triangulation: compute intersection of two backprojected rays)实现重建 射影重建 理想情况下 真实情况下 Linear Triangulation methods SVD 几何误差 possibility to compute using LM (for 2 or more points) or directly (for 2 points) Reconstruction ambiguity:projective The projective reconstructiontheorem If a set of point correspondences in two views determine the fundamental matrix uniquely, then the scene and cameras may be reconstructed from these correspondences alone, and any two such reconstructions from these correspondences are projectively equivalent Find x,x’ Find F Find P,P’ Find X Stratified reconstruction(分层重建) Projective reconstruction Affine reconstruction Metric reconstruction Affine reconstruction 仿射不变量： 保持无穷远平面不变。反之，亦然。 保持平行性。 保持物体的体积比、平行图形（或共面图形）的面积比、平行线段（或共线线段）的长度比不变 12DOF Affine reconstruction But how do we obtain so as to achieve affine reconstruction? An affine reconstruction can be sufficient depending on applications, e.g. calculate mid-point, centroid, parallelism Translational motion (e.g. in a car on a straight road, moon) Points at infinity are fixed for a pure translation reconstruction by those xi? xi which are on ,we can obtain the and affine reconstruction Scene Constraints Find 3 points which are known to lie on the plane at infinity 3 sets of parallel lines allow to uniquely determine 找三组平行线，求得V1,V2,V3 求得 最后 A vanishing point v and its corresponding line l’ ([v]xP)X=0 and (l’TP’)X=0 Scene Constraints Scene Constraints Distance ratios on a line 交比不变性 Scene constraints –infinity homography Reconstruction ambiguity:similarity（＝Metric and Euclidean reconstruction） Metric Reconstruction 7DOF 相似不变量： 保持绝对二次曲线不变。反之，亦然。 保持平行性。 保持物体的体积比、平行图形（或共面图形）的面积比、平行线段（或共线线段）的长度比不变 保角 Key to affine reconstruction: plane at infinity Key to metric reconstruction: absolute conic Metric Reconstruction 射影变换 仿射变换 刚体变换 交比不变，共线、共面性不变 平行线变为平行线，