考虑宽基线立体像对视角变化的航空遥感影像密集匹配方法

Dense Matching Method of Aerial Remote Sensing Image Considering the Change of View Angle of Wide Baseline Stereo Image

针对宽基线摄影条件下,立体像对视角变化以及光照条件变化等因素造成的航空遥感影像密集匹配困难问题,提出一种有效的航空遥感影像密集匹配方法。该方法利用尺度不变特征变换(SIFT)算法,计算基本矩阵和单应矩阵,获取航空遥感影像局部特征,并构建复杂光照条件下的彩色SIFT算法。在此基础上,采用支持向量回归(SVR)方法来实现宽基线图像密集匹配。其中在SVR算法的基础上加入一些约束条件,生成更为精确的匹配点,并通过SVR算法的扩散方式可以得到较为密集的匹配结果,完成初始特征匹配。利用拓扑过滤器的投票机制匹配大量SIFT特征,并剔除错误匹配项,获得高准确率的初始匹配作为训练数据,实现基于高准确率初始匹配的为SVR构建,并给出的预测搜索新匹配。实验结果表明,该方法可在宽基线条件下对航空遥感影像实施正确密集匹配,获取规模相当的密集匹配点,且密集匹配可视效果良好。

In view of the difficulty of dense matching of aerial remote sensing image caused by the change of stereo image pair angle and the change of light condition under the condition of wide baseline photography,an effective method of dense matching of aerial remote sensing image is proposed.In this method,the scale invariant feature transform(SIFT)algorithm was used to calculate the basic matrix and the homography matrix,obtain the local features of the aerial remote sensing image,and construct the color SIFT algorithm under complex illumination conditions.On this basis,support vector regression(SVR)was used to achieve wide baseline image dense matching.Based on the SVR algorithm,some constraints were added to generate more accurate matching points,and through the diffusion of SVR algorithm,more intensive matching results can be obtained to complete the initial feature matching.The voting mechanism of topology filter was used to match a large number of SIFT features,and the wrong matching items were eliminated.The initial matching with high accuracy is obtained as training data.The SVR based on the initial matching with high accuracy was constructed,and the new matching of prediction search was given.The experimental results show that this method can achieve the correct dense matching of aerial remote sensing image under the wide baseline condition,and obtain the dense matching points with the same scale,and the dense matching visual effect is good.