基于网格对应的双约束特征点匹配算法

常用的特征点匹配算法通常设置严苛的阈值以剔除错误匹配,但这样也会导致过多的正确匹配被删除。针对这一问题,提出了一种采用双约束的特征点匹配方法。首先,在局部上统计特征点匹配数量,运用网格对应的方法过滤部分错误匹配;然后,在全局上运用RANSAC方法计算基础矩阵,通过极线约束对匹配进行再一次筛选。实验表明,相比于传统的匹配算法,该算法能在不增加算法运行时间的前提下,获得更高数量和更高质量的匹配集合。

Feature Point Matching Based on Double Constraints

Commonly used feature point matching algorithms usually set strict thresholds to eliminate false matches,which may cause many correct matches to be deleted.To overcome this problem,a feature matching algorithm using double constraints isproposed.Firstly,we statistically count the number of feature point matches in local to establish a grid correspondence,which can be used to filter out partial false matches.Then,the RANSAC was globally introduced to calculate the fundamental matrix,and the matching is once again filtered by the epipolar constraint.Experiments show that compared to the traditional matching algorithm,our algorithm can obtain a higher number and higher quality matching set without additional running time.