The extraction and use of image planes for three-dimensional metric reconstruction

摘要: The three-dimensional (3D) metric reconstruction of a scene from two-dimensional images is fundamental problem in Computer Vision. major bottleneck the process retrieving such structure lies task recovering camera parameters. These parameters can be calculated either through pattern-based calibration procedure, which requires an accurate knowledge scene, or using more flexible approach, known as autocalibration, exploits point correspondences across images. While presence object, autocalibration constraints are often cast into nonlinear optimization problems sensitive to both image noise and initialization. In addition, fails for some particular motions camera. To overcome these problems, we propose combine address this thesis (a) extracting geometric information uncalibrated images, (b) obtaining robust estimate affine camera, (c) upgrading refining one. particular, method identifying planar structures another recognize parallel pairs planes whenever available. identified then used obtain 3D without resorting traditional error prone calculation vanishing points. We also refinement which, unlike existing ones, capable simultaneously incorporating plane parallelism perpendicularity process. Our experiments demonstrate that proposed methods provide satisfactory results.