Abstract: Aiming at the problem that the variation of wide haseline image view makes it difficult to obtain the raliable matching results by the exisiting linear feature matching methods. A local structure constraint based robust line segment matching method is proposed in this paper for wide-baseline images with viewpoint change. Firstly, extract the line segment in images and grouped according to the rules. Secondly, viewpoint invariant feature region is constructed based on the geometric relationship between line segments, a feature descriptor is computed for each feature region and initial matches are generated by measuring feature descriptor similarity. Then, an outlier elimination method is designed by exploring the relationship of feature region vertices coordinates of the initial matches. Finally, a line segment matching expansion step is proposed to improve matching performance. The experimental results demonstrate that the proposed method performs better than traditional methods for wide-baseline images with large viewpoint change in terms of a large number of correct matches and matching precision.

Key words: wide-baseline images, line segment matching, viewpoint variation, local structure constraint, expansion matching