地球信息科学学报 ›› 2020, Vol. 22 ›› Issue (9): 1766-1778.doi: 10.12082/dqxxkx.2020.190334

• 地球信息科学理论与方法 • 上一篇    下一篇

基于直线间几何属性特征相似性约束的组直线匹配算法

何腕营1(), 王竞雪1,2,*()   

  1. 1.辽宁工程技术大学 测绘与地理科学学院,阜新 123000
    2.西南交通大学 地球科学与环境工程学院,成都 611756
  • 收稿日期:2019-06-25 修回日期:2019-10-23 出版日期:2020-09-25 发布日期:2020-11-25
  • 通讯作者: 王竞雪 E-mail:172862389@qq.com;xiaoxue1861@163.com
  • 作者简介:何腕营(1994— ),女,辽宁沈阳人,硕士生,主要从事直线匹配、三维重建理论与方法研究。E-mail: 172862389@qq.com
  • 基金资助:
    国家自然科学基金项目(41871379);地球观测与时空信息科学国家测绘地理信息局重点实验室开放基金项目(201801)

Pair-wise Line Matching Algorithm based on Feature Similarity Constraints of Geometric Attributes between the Lines

HE Wanying1(), WANG Jingxue1,2,*()   

  1. 1. School of Geomatics, Liaoning Technical University, Fuxin 123000, China
    2. Faculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu 611756, China
  • Received:2019-06-25 Revised:2019-10-23 Online:2020-09-25 Published:2020-11-25
  • Contact: WANG Jingxue E-mail:172862389@qq.com;xiaoxue1861@163.com
  • Supported by:
    National Natural Science Foundation of China(41871379);Key Laboratory of Earth Observation and Geospatial Information Science of NASG(201801)

摘要:

当影像中存在多个相同或相近的直线特征时,仅利用单直线特征间的相似性进行匹配容易导致算法失效,因此本文提出基于直线间几何属性特征相似性约束的组直线匹配算法。该算法利用直线间的拓扑关系分别对2幅影像上提取的直线进行编组得到特征直线组,并将其作为匹配基元;然后利用核线约束确定候选同名直线组的搜索范围,精简了候选直线组的数量;依据直线间几何属性特征向量的仿射不变性建立直线组的匹配关系,将仿射不变量交比作为基础几何不变性测度,并延伸设计仿射相似度参数,计算出目标直线组与每个候选直线组的仿射相似度或一般相似度,确定2种相似度下总体相似度值最大的特征直线组为其同名直线组,最后将同名直线组分裂为2对同名单直线,对分裂后的结果进行整合可以显著降低冗余匹配,得到“一对一”的匹配单直线。为了验证算法的可靠性,实验选取网上公开的5组典型近景影像进行测试,通过与其他匹配算法的对比分析,结果表明该算法应对影像间存在的视角、旋转和尺度变换等复杂条件均取得了较高的匹配精度,匹配正确率最高有14.5%的提升,且阈值的选择对匹配结果影响微弱,验证了基于直线间几何属性特征相似性约束的组直线算法鲁棒性和匹配稳定性较强。

关键词: 直线组匹配, 核线约束, 直线几何属性, 仿射不变性, 特征向量, 特征相似性约束, 方向约束

Abstract:

Many same or similar line features may hinder line matching in the image, matching only based on the similarity between individual line, which easily lead to the failure of the algorithm. Therefore, this paper proposed a reliable pair-wise line matching algorithm based on feature similarity constraints of geometric attributes between the lines. There are two challenges in constructing robust feature similarity constraints based on pair-wise line matching. The first challenge is to generate grouped line pairs under unstable lines extraction. This algorithm is handled by the basic geometric relationships such as distance and angle between the lines. The second challenge is to design a reliable feature descriptor robust to large viewpoint changes taking into account that the line pairs may not be coplanar and their endpoints are inaccurate. This algorithm salient not only against a range of viewpoint changes for close-range image but also large affine transformation. The construction method is described as follows. Firstly, line segments are extracted by using a line segment detector method and the corresponding points obtained by using SIFT matching points in the reference image and searching image.The algorithm generates grouped line pairs from lines extracted from the reference image and searching image according to the basic geometric relationships such as distance and angle between the lines, and it takes the grouped line pairs as matching primitives. Then the method employs the epipolar constraint to evaluate candidate line pairs. Line matching algorithm based on geometric attributes of lines for descriptor and similarity measure of line pairs are presented which is more distinctive by describing the relationship between every two pairs. The configuration of two line pairs is described by distinguishing two cases, in the first case, the similarity of the two line pairs is called as affine similarity; In another case, it is called general similarity. The affine similarity and general similarity are used as the overall similarity measure in this algorithm and are determined based on the calculated feature vectors between the target line pairs and each candidate line pairs. Instead of screening all candidate line pairs, the optimal line pairs mapping that maximizes the similarity measure between the two line pairs. Finally, the direction constraint which can provide a solution for angle transformation caused by image rotation in line pairs matching is used to perform the correspondence of individual line, it resolves the corresponding line pairs into two pairs of corresponding individual lines, and obtains one to one matching results after the post-processing of checking. Five typical groups of close-range image pairs with angle, rotation, and scale transformation are used as the experimental dataset, which is used to complete the line pairs matching experiments by the proposed algorithm. In comparison with other line matching algorithms, the proposed method can obtain more accurate line matching results in different typical close-range image pairs, and its matching ratio increased by 14.5. The experiment results demonstrate that the effect of threshold selection is weak and the algorithm is robust which achieves reliable line matching results.

Key words: pair-wise line matching, epipolar constraint, geometric attributes of lines, affine invariance, feature vectors, feature similarity constraint, direction constraint