基于相对定向和三角形约束的近景影像匹配

摘要

A new image matching method is introduced based on triangle constraint into the process of automatic relative orientation of close-range imaging considering the large proportion of outliers . Firstly, this method used an algorithm for Harris corner detection to extract the keypoints,then the RANSAC method and pope-hinsken algorithm were led into the process of relative orientation to construct the epipolar geometry after the initial matching. Secondly, it triangulated the well-defined inliers based on the Delaunay triangle criterion ,then extracted the new keypoints using a smaller threshold in pair of images. Lastly, it integrated the triangle constraints to acquire new matching points, by means of relative orientation to get the new inliers , then updated the constraint triangulation , until it got enough inliers. The experimental result proves the proportion of outliers is reduced availably,and the reliability of image matching is improved obviously.%针对近景影像自动相对定向过程中初匹配点对中误匹配率较大的情况,将基于三角形约束方法引入近景影像匹配.首先利用Harris算子提取特征点集,经初匹配后运用基于RANSAC方法和P-H算法的相对定向构建核线几何得到内点集,删除误匹配点生成同名Delaunay三角网；在同名三角形的约束下通过缩小Harris特征点阈值得到新内点集,实时插入同名三角网中,并不断动态更新直到生成足够数量且分布均匀的内点.实验结果表明,文中所提出的基于相对定向和三角形约束的近景影像匹配方法有效减小了误匹配率,提高了匹配可靠度.