基于改进的SIFT算法的工件轮廓配准方法  

Workpiece Contour Registration Method Based on Improved SIFT Algorithm

作　　者：林少鑫 方千山[1] LIN Shaoxin;FANG Qianshan(College of Mechatronics and Automation,Huaqiao University,Xiamen 361021,China)

机构地区：[1]华侨大学机电及自动化学院,福建厦门361021

出　　处：《机械工程师》2022年第10期4-7,共4页Mechanical Engineer

摘　　要：工件轮廓的配准是机器人视觉引导系统中对工件位姿感知的重要组成部分,通过工件轮廓与模板轮廓的配准可以获得工件的位姿信息,从而引导机器人进行抓取和安装。文中针对图像配准算法中特征提取的SIFT算法计算复杂、特征提取效率慢的缺点,在工件轮廓特征提取过程中,直接在原图像上计算高斯尺度空间图像,不进行上下采样,最后在高斯差分图像上寻找极值点作为轮廓特征点,最后通过与模板提取的特征点进行逐个匹配,按照重合度进行排序,选择重合度最高的特征点计算仿射变换的参数。实验表明,改进的SIFT算法能够完成工件轮廓图像与模板轮廓图像的配准,并将特征提取速度提高了近10倍。The registration of the workpiece contour is an important part of the perception of the workpiece pose in the robot vision guidance system.Through the registration of the workpiece contour and the template contour,the pose information of the workpiece can be obtained,thereby guiding the robot to grasp and install.The SIFT algorithm for feature extraction in the image registration algorithm has the disadvantages of complicated computation and slow feature extraction efficiency.In the process of workpiece contour feature extraction,this paper directly calculates the Gaussian scale space image on the original image without up-sampling,and searches for extreme points on the Gaussian difference image as contour feature points.Finally,the feature points extracted from the template are matched one by one,sorted according to the degree of coincidence,and the feature points with the highest degree of coincidence are selected to calculate the parameters of the affine transformation.Experiments show that the improved SIFT algorithm can realize the registration of the workpiece contour image and the template contour image,and improve the feature extraction speed by nearly 10 times.

关 键 词：机器视觉 轮廓配准 视觉引导 SIFT 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]