基于粗定位-精匹配的双目视觉目标定位方法  

Binocular Vision Target Location Method Based on Coarse Location and Fine Matching

作　　者：杜玉晓[1] 郑晓森 陈祎杭 王小桥 王公东 钟楠 Du Yuxiao;Zheng Xiaosen;Chen Yihang;Wang Xiaoqiao;Wang Gongdong;Zhong Nan(College of Automation,Guangdong University of Technology,Guangzhou 510006,China;Hisense Electronics Co.,Ltd.Jiangmen529000,China)

机构地区：[1]广东工业大学自动化学院,广东广州510006 [2]广东海信电子有限公司,广东江门529000

出　　处：《自动化与信息工程》2021年第5期11-17,共7页Automation & Information Engineering

摘　　要：为提高双目视觉系统定位精度,提出一种基于粗定位-精匹配的双目视觉目标定位方法。利用Canny边缘检测算法对左、右图像中的目标物体进行识别,提取目标物体最大、最小矩形区域,从而实现目标物体的粗定位。采用SIFT算法得到左、右图像中目标的特征点;用RANSAC得到精确匹配点,并计算出左、右图像中物体之间的透视变换关系;根据透视变换关系确定左图像的目标特征点在右图像目标潜在匹配点的位置,得到目标物体的三维位姿及质心坐标,从而实现目标物体的精匹配。实验结果表明:本文方法平均测距误差为4.6 mm,平均耗时为1.265 s,相比SIFT、SURF特征匹配定位方法,具有定位精度高、运行时间短的特点。In order to improve the localization accuracy of binocular vision system,a binocular vision target localization method based on coarse localization fine matching is proposed.Canny edge detection algorithm is used to recognize the target object in the left and right images,and the maximum and minimum rectangularregions of the target object are extracted,so as to realize the rough positioning of the target object.SIFT algorithm is used to obtain the feature points of the target in the left and right images;The exact matching points are obtained by RANSAC,and the perspective transformation relationship between the objects in the left and right images is calculated;According to the perspective transformation relationship,the position of the target feature point of the left image at the potential matching point of the target in the right image is determined,and the three-dimensional pose and centroid coordinates of the target object are obtained,so as to realize the fine matching of the target object.The experimental results show that the average ranging error of this method is 4.6 mm and the average time is 1.265 s.Compared with sift and surf feature matching positioning methods,this method has the characteristics of high positioning accuracy and short running time.

关 键 词：双目立体视觉 粗定位 精匹配 SIFT算法 RANSAC 透视变换 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]