双目视觉是机器视觉中经典有效的方法。针对无色差光滑曲面特征点难以提取的问题,提出一种用均匀激光网格创造曲面上色彩特征的方法,以及特征点提取和匹配算法。首先,提取激光网格线,采用膨胀与腐蚀等算法将激光线断点拼接并转化为单像素连通线;然后,利用连通线上点的八邻域及四邻域的连通性对特征点进行提取;最后,对提取后的特征点根据位置信息进行编号,完成特征点的匹配。实验结果表明,该算法稳健性好,提取精度高,匹配正确率及重复率高,三维重建后特征点间的距离误差小于0.05 mm、标准差为0.0362 mm。该算法直接对特征点进行提取及匹配,与离散特征线获取特征点的方法相比精度更高,可应用于曲面上点的均匀抽样检测。

Binocular vision is a classical and efficient method in machine vision. Aiming at the problem that it is difficult to extract feature points of smooth surfaces without chromatic aberration, we propose a method of creating color features on surfaces with uniform laser grids, and a feature point extraction and matching algorithm. Firstly, in order to extract the laser grid line, the laser line breakpoint is spliced and transformed into a single pixel connected line by the expansion and corrosion algorithm. Then, the feature points are extracted by the connectivity of eight neighborhood and four neighborhood of points on the connected line. Finally, the extracted feature points are numbered according to the location information to complete the matching of feature points. The experimental results show that the algorithm has good robustness, high extraction accuracy, high matching accuracy and high repetition rate. The distance error between the feature points after three-dimensional reconstruction is less than 0.05 mm, and the standard deviation is 0.0362 mm. The proposed algorithm can carry out feature point extraction and matching directly, which is more accurate than the method of discrete feature lines as feature points. It can be applied to the uniform sampling detection of points on surfaces.