几何约束下曲面场景的立体匹配迭代算法

Feature matching iteration algorithm based on geometric constraints in stereo views of curved scenes

使用对极约束和同形映射两种几何约束，通过采用面积检测、一致性约束误差检测、概率筛选规则和对称优化等策略，成功地实现了曲面场景图像的特征点匹配。该方法未使用任何与灰度相关的信息和与场景相关的约束，极大地提高了方法的实用性和鲁棒性。整个匹配过程快速有效，并在不同的弱标定关系已知的真实图像数据上成功地得到实现。

Area detection, consistent constraint error detection, probability selection rule and symmetry optimization were suecessfully applied to the correction and completion of feature matching of curved scenes by epipolar constraints homography mapping in curved scenes. Only geometric constraints rather than scene-dependent constraints were involved in the whole matching process, so the method had great robustness and practicality. The results for experiments with real image data are given in the paper.