近景影像三角网内插点密集匹配方法

Dense matching method of inserting point into the Delaunay triangulation for close-range image

针对目前密集匹配中依据种子点寻找新特征点存在计算复杂的问题,该文提出一种近景影像Delaunay三角网内插点密集匹配方法。该方法首先采用尺度不变特征变换算子匹配特征点,通过随机抽样一致性算法对特征点进行优化,以获取高精度同名点;依据同名点构建Delaunay三角网,在同名相似三角形内,以内插重心点作为匹配基元,并对内插点进行色彩信息相似性约束和极线约束,剔除粗差提高匹配结果精度;在匹配传播过程中,新特征点不断插入三角网中,对三角网进行动态更新,用于约束后续匹配。该方法能够避免繁琐计算,同时具有较高的可靠性,适用于不同类型的近景影像数据。

Aiming at the computational complexity problem of searching new feature points based on seed points in dense matching, this paper presented dense matching method of inserting point into the Delaunay triangulation for close-range image. Firstly, SIFT matching and RANSAC algorithm was used to obtain highly accurate corresponding points, Delaunay triangulation was constructed by using correspond- ing points secondly, inserted the center of gravity points into the same triangle as matching element using color similarity constraint and epipolar constraint; finally, achieved precise matching purposes. In the process of matching transmission, new feature points were inserted into the Delaunay triangulation, updated the Delaunay triangulation dynamically for constraining the subsequent matches. This method could avoid complicated calculations with high reliability, and it was suitable for different types of close-range images.