基于神经网络的双目远心镜头标定方法研究  

Research on binocular telecentric lens calibration method based on neural network

作　　者：梁天为 朱呈祥 陈浩 Liang Tianwei;Zhu Chengxiang;Chen Hao(School of Electrical Engineering&Automation,Jiangsu Normal University,Xuzhou 221116,Jiangsu,China;Matrix Time(Shanghai)Co.,Ltd,Shanghai 201100,China)

机构地区：[1]江苏师范大学电气工程及自动化学院,江苏徐州221116 [2]聚时科技(上海)有限公司,上海201100

出　　处：《江苏师范大学学报（自然科学版）》2020年第4期75-78,共4页Journal of Jiangsu Normal University：Natural Science Edition

基　　金：江苏省高校自然科学基金资助项目(16KJB120001);徐州市科技计划项目(重点研发KC17144);江苏师范大学研究生科研与实践创新计划项目(2020XKT095)。

摘　　要：利用BP神经网络方法对双目远心镜头标定,无需建立远心镜头相关的数学模型及求解相机的内外参数,不仅可以缩小利用普通镜头标定的高度误差,而且还有效避免了传统相机标定的繁琐.利用一种圆形特征提取算法获得标靶左右图像匹配特征点的二维像素坐标后,再用左右相机所拍摄的标定图像进行学习训练,获取左右图像特征点二维像素坐标和三维世界坐标之间的映射关系.实验结果表明,该方法的样本标定点物理坐标误差小于1μm,测量区域内的随机匹配点物理坐标最大误差为9.8μm,能够以工业自动化形式实现高精度的双目远心镜头标定.BP neural network is used to calibrate binocular telecentric lens.It is not necessary to establish the mathematical model of telecentric lens and solve the camera s internal and external parameters,it can not only reduce the height error of calibration with ordinary lens,but also avoid the tedious calibration of traditional camera effectively.A circular feature extraction algorithm is used to obtain the 2D pixel coordinates of the matching feature points of the left and right images of the target,and then the calibrated images taken by the left and right cameras are trained,the mapping relationship between the two-dimensional pixel coordinates and the three-dimensional world coordinates of the left and right image feature points is obtained.The experimental results show that the physical coordinate error of the sample points is less than 1μm,and the maximum physical coordinate error of the random matching points in the measuring area is 9.8μm,high precision binocular telecentric lens calibration can be realized in the form of industrial automation.

关 键 词：相机标定 远心镜头 BP神经网络 双目视觉 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]