相对定向中错误匹配的剔除方法研究

The Research on False Matches Filtering of Automatic Relative Orientation

数字影像的全自动相对定向已不再仅仅利用传统的6点而是利用相当多的点来完成,这样就可以有大量的多余观测以提高定向参数解算的可靠性。然而,匹配点集中不可避免地会存在错误匹配,影响定向参数的准确性,如何剔除这些错误匹配是一个很重要的问题。本文简要介绍了全自动相对定向的流程,阐明误匹配剔除在全自动相对定向过程中所起的作用,根据最小中值平方法的思想原理,提出基于RANSAC的核线约束和仿射变换约束的误匹配剔除流程。实验结果表明,算法能有效滤除误匹配,保证了相对定向结果的正确性。

The full-automatic relative orientation of digital images makes use of quite a number of points instead of traditional 6 points and plentiful redundant observations can improve the reliability of calculating orientation parameters. However, inevitable mismatch due to aggregation of matching points influences the accuracy of orientation parameters. Then, how to reject this kind of mismatch is an important problem. This paper introduced the workflow of full-automatic relative orientation, illustrated the roles of mismatch rejection in the process of full-automatic relative orientation and proposed a new algorithm to reject mismatch based on epipolar constrains and affine transformation with RANSAC according to the least median square method. The experimental results show the proposed algorithm can effectively filter mismatch and ensure the accuracy of relative orientation.