中文摘要：

针对三维外形测量以及表面位移、损伤、应变测量中的随机纹理图像,提出了一种特征匹配算法。该算法首先结合H arris角点探测和一个72维的角点特征描述向量,快速获取随机纹理图像对之间的粗匹配像点集合,然后通过灰度相关约束和视差梯度约束对粗匹配集合进行检验,剔除粗匹配集中的错误匹配。通过在误匹配剔除的过程中采用与粗匹配阶段相关联的自适应约束阈值,从而使最终的匹配结果不敏感地依赖于阈值参数的选取,而是在一个较大的取值范围内都能使最终结果在严格控制误匹配的前提下,得到足够多的正确匹配点对。通过实验对提出的方法进行了验证。

英文摘要：

A feature matching algorithm is put forward for the random texture images applied in 3D surface shape measurement and surface displacement,damage and strain measurement.Firstly,by combining the Harris corner detector and a 72-dimensional feature descriptor,a coarse matching set of the random texture images is rapidly obtained.Then,both the intensity correlation limit and the disparity gradient constraint are imposed to filter out false correspondences from the matching set.Due to using an adaptive constraint threshold related to the parameter used for generating the coarse matching set,the final matching result is insensitive to the threshold selection.Consequently,sufficient correct correspondences can be collected in a wide parameter range,while false correspondences can be strictly controlled.The method is verified by experiments.