基于倾斜校正和视差估计的遥感影像稠密匹配方法

针对遥感影像稠密匹配效率较低、匹配精度和密度不够高的缺陷,提出了一种基于倾斜校正和视差估计的遥感影像稠密匹配方法。首先基于稀疏同名点、通过倾斜校正生成核线影像;利用稀疏同名点构造三角网,以三角形为插值单元,预测核线影像对中各像素的初始视差;在初始视差的邻域内沿核线一维搜索最佳匹配点,搜索时增量计算匹配窗口的灰度均值,且将搜索步长设为2,以提高搜索效率。实验结果表明,本文方法可以达到逐像素的匹配密度,且匹配的正确率和精度很高;基于稠密点重建的真实感三维场景证明了本文方法的有效性和实用性。

Dense Correspondence Approach for Remote Sensing Imagery Based on Tilt Correction and Disparity Estimation

In view of the defects of high error rate and low efficiency of dense correspondence of aerial remote sensing imagery, an dense correspondence approach for aerial remote sensing imagery based on tilt correction and parallax estimation was proposed. First, the epipolar imagery was generated by tilt correction based on the sparse corresponding points. Afterwards, the initial disparity of the epipolar image couple was interpolated from the sparse correspondence, and the best correspondence was searched around the nearby area of initial parallax along the epipolar. In order to improve the search efficiency, the average gray value of the matching window was computed incrementally, and the search step was set to 2 during the search process. Experimental result illustrates the matching density of pixel by pixel can be achieved, and the accuracy rate and precision of the matching result are high. The realistic 3D scene reconstructed from the dense corresponding points demonstrates the effectiveness and practicability of the proposed approach.