核线驱动与自适应窗口相结合的建筑物角点稳健匹配算法

针对景深突变、区域遮挡等因素导致的倾斜立体影像中同名建筑物角点难以准确匹配的难题,提出一种基于核线驱动和自适应窗口的鲁棒匹配算法。算法分3个阶段:1提取重复度较高、分布均匀的角点特征,并对立体像对进行多类型互补仿射不变特征匹配,以用于后续的几何关系估计;2利用随机采样一致性(random resample consensus,RANSAC)算法来排除地面区域的角点特征,并采用改进的RANSAC算法估计两组倾角迥异的核线关系;3基于步骤2估计的双核线关系求取景深突变区域中角点的近似几何单应变换,并联合自适应窗口方法实现同名建筑物角点的精确匹配。最后,通过多组高分辨率无人机倾斜影像数据验证了该算法的有效性与优越性。

A Robust Matching Algorithm for Corner Points of Building Based on Epipolar Driving and Self-Adaptive Window

Some objective factors,namely discontinuity depth of field,region occlusion and geometric distortion,make matching a tough work for conjugate corner points of building within oblique stereo images,to resolve this problem,a robust matching algorithm is proposed based on epipolar driving and self-adaptive window.The algorithm was involved in three stages:(1)corner points with high recall rate and uniform distribution were detected,and complementary affine invariant feature matching was implemented on image pairs for the purpose of estimating geometric transformation;(2)the corner points located ground area would be removed by random resample consensus algorithm,and two-group epipolar relationship with big separative angle were computed based on modified algorithm of random resample consensus;(3)the homograph transformation between corresponding corner points with discontinuity depth of field can be approximately calculated based on the two-group epipolar result of the second stage,and further integrated the method of self-adaptive window,the accurate matching corner pints of building were obtained.Experiments on multi-group oblique images taken by unmanned aerial vehicle (UAV)demonstrate the effectiveness of the proposed algorithm.