基于相位靶和神经网络的单目相机标定  

Monocular camera calibration based on phased target and neural network

作　　者：张党成 董秀成 雎雅玲 向贤明 ZHANG Dangcheng;DONG Xiucheng;JU Yaling;XIANG Xianming(School of Electrical Engineering and Electronic Information Xihua University,Chengdu 610039,China;Sichuan University Jinjiang College,Meishan 620860,China)

机构地区：[1]西华大学电气与电子信息学院,四川成都610039 [2]四川大学锦江学院,四川眉山620860

出　　处：《光学技术》2024年第3期346-353,共8页Optical Technique

基　　金：国家自然科学基金(11872069);四川省中央引导地方科研发展专项(2021ZYD0034)。

摘　　要：针对传统基于BP神经网络的标定方法标定精度低,深度方向误差大等问题,提出了一种基于相位靶和径向基函数(RBF)神经网络的单目相机标定方法。采用三步相移法的特征提取方法,并使用多频法计算绝对相位,将相位靶特征点承载的绝对相位转换到三维空间,建立特征点图像坐标与世界坐标之间的对应关系,最后使用RBF神经网络完成二维图像坐标到三维空间坐标的直接映射。实验结果表明,对比传统使用BP神经网络进行标定与棋盘格,圆形标定靶的标定结果,该方法的平均标定误差为0.0980mm,同时在焦距1.4mm,视场220°鱼眼镜头下,仍能保持较高的精度,证明了所提方法的可行性和有效性。Aiming at the problems of low calibration accuracy and large depth direction error of traditional calibration methods based on BP neural network,a monocular camera calibration method based on phase target and Radial basis function(RBF)neural network is proposed.The feature extraction method adopts a three-step phase shift method,and the multi frequency method is used to calculate the absolute phase.The absolute phase carried by the phase target feature points is converted into three-dimensional space,and the corresponding relationship between the feature point image coordinates and world coordinates is established.Finally,the RBF neural network is used to directly map the two-dimensional image coordinates to the three-dimensional space coordinates.The experimental results show that compared with the traditional use of BP neural network for calibration and the calibration results of checkerboard and circular calibration targets,the average calibration error of this method is 0.0980mm,At the same time,under a focal length of 1.4mm and a 220°field of view fisheye lens,high accuracy can still be maintained,proving the feasibility and effectiveness of the proposed method.

关 键 词：相位靶 RBF神经网络 绝对相位计算 相机标定 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术] TB811[自动化与计算机技术—计算机科学与技术]