基于迭代式动态规划的影像密集匹配  

Dense Image Matching based on Iterative Dynamic Programming

作　　者：廖小超 胡坚[1] 李传荣[1] 吴克[3] 薛博[3] 李群智[3] 

机构地区：[1]中国科学院定量遥感信息技术重点实验室,北京100094 [2]中国科学院大学,北京100049 [3]中国空间技术研究院,北京100094

出　　处：《遥感技术与应用》2015年第1期92-98,共7页Remote Sensing Technology and Application

摘　　要：针对动态规划影像密集匹配中因匹配的整体相关性导致的误匹配点连带扩散效应问题,提出了一种基于迭代式动态规划的影像密集匹配算法。该算法在影像密集匹配过程中引入了视差方向一致性、视差突变性作为迭代判定准则,通过对动态规划匹配结果中不满足迭代条件的候选匹配点子集进行分析,在候选匹配点子集中识别出误匹配点并去除之,反复迭代直至满足迭代准则,从而解决了误匹配点的连带扩散效应问题。该算法已成功应用于嫦娥三号遥操作项目,经在轨应用的检验表明,提出的基于迭代式动态规划的影像密集匹配算法能够极大地降低立体影像匹配中密集同名点的误匹配率。A dense image matching method based on iterative dynamic programming is proposed to solve the mismatching points diffusion effect caused by the overall integrity in the dynamic programming dense image matching.The parallax direction consistency and parallax mutation are introduced as an iterative decision criterion in the algorithm.By analyzing a set of mismatching point candidates which are against iterative condition in dynamic programming matching result,the mismatching points are identified and removed from the set.The iteration continues until the iterative decision criterion is fulfilled and the mismatching points diffusion effect will be solved.The algorithm proposed in the paper has been successfully applied to Chang'e-3project.The results in orbit show that the dense imagery matching algorithm based on iterative dynamic programming is able to greatly reduce the mismatch rate of corresponding image points in stereo image match.

关 键 词：影像匹配 动态规划 代价函数 密集匹配 迭代式动态规划 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]