几何特征约束的SIFT特征匹配改进算法  

Improved SIFT Feature Matching Algorithm with Geometric Feature Constraints

作　　者：张喜民 詹海生[2] 余奇颖 ZHANG Xi-min;ZHAN Hai-sheng;YU Qi-ying(School of Information Engineering,Shaanxi Institute of International Trade&Commerce,Xi'an,Shaanxi 712046,China;School of Computer Science and Technology,Xidian University,Xi'an,Shaanxi 710071,China)

机构地区：[1]陕西国际商贸学院信息工程学院,西安陕西712046 [2]西安电子科技大学计算机学院,西安陕西710071

出　　处：《计量学报》2023年第8期1182-1187,共6页Acta Metrologica Sinica

基　　金：陕西省自然科学基础研究计划(2021JM-539)。

摘　　要：图像配准精度直接影响机器视觉尺度检测的精度,针对尺寸不变特征变换(SIFT)特征匹配未考虑特征点之间的几何关系,对于小插件等灰度变化较平滑的图像易产生误匹配点的问题,提出了几何特征约束的SIFT特征匹配改进算法。首先,用基于边界拓扑分析的图像跟踪算法提取目标轮廓;然后,用轮廓特征对SIFT算法进行约束并利用随机抽样一致(RANSAC)算法去除“离群点对”;最后,估计变换矩阵并完成图像精确配准。基于高分辨率工业相机和高性能计算机构建精密尺度检测系统,以手机USB接口插件为对象进行实验,实验结果表明:该算法图像配准精准度可达95.31%,与SIFT算法、SIFT+RANSAC算法相比较,配准精准度得到了较大提高,该算法已应用于某检测设备研制中。The accuracy of image registration directly affects the accuracy of machine vision dimension detection.Aiming at the fact that scale-invariant feature transform(SIFT)feature transform algorithm was less involved the geometric relationship between feature points,and it was easy to generate mismatch points for the capture image with smooth varied grayscale,such as small widget,an improved SIFT feature matching algorithm with geometric feature constraints was proposed.Firstly,the target contour profile was extracted by the image tracking algorithm based on boundary topology analysis.Secondly,the SIFT algorithm was constrained by geometric contour profiles of images,and the Random sample consensus(RANSAC)algorithm was taken to remove outlier point pairs.Finally,image accurate registration transformation was performed by calculated transformation matrix.A machine vision precision detection and measurement system were developed based on high-resolution industrial camera and high-performance computer,and experiments were carried out on the USB interface plug-in parts of mobile phone as samples.The experimental results showed that the accuracy of image registration of this algorithm can reach 95.31%.Compared with SIFT algorithm and SIFT+RANSAC algorithm,the registration accuracy was greatly improved.The algorithm has been applied to develop an automation dimension detection and measurement system.

关 键 词：计量学 机器视觉 图像配准算法 尺寸不变特征变换 特征匹配 边界跟踪 随机抽样一致算法 

分 类 号：TB96[机械工程—光学工程]