基于立体靶标的摄像机标定方法

Camera calibration method based on 3D board

为获取视觉测量系统中二维图像到三维空间位置的变换关系,提出一种基于立体靶标的摄像机标定方法.针对无畸变小孔成像模型,使用最小二乘法求解初始投影矩阵后通过LM准则对其优化;根据多张图像对应的投影矩阵,求解摄像机内参数及各相应外参数;引入二阶径向畸变模型,建立理想图像坐标和实际图像坐标间的方程求解初始畸变系数;使用LM准则全局优化,得到更精确的摄像机内外参数及畸变因子.实验结果表明:仿真图像数据中高斯噪声小于0.5像素时,摄像机等效焦距误差小于0.1%,图像主点误差小于0.5像素;在相同噪声等级下,标定使用图像数越多获得的参数标准差越小;该方法标定参数对应的位置残差小于其他立体靶标标定方法.该标定方法具有较高的标定精度,且增加标定图像数有助于抑制噪声获取稳定的摄像机参数.

A method of camera calibration based on 3D calibration board,which is used for vision measuring system,is proposed in this paper.Transform matrix is worked out with least square method on an ideal pin-hole camera model and then optimized by Levenberg-Marquardt（LM） algorithm.According to several transform matrices,initial camera intrinsic parameter matrix and corresponding extrinsic parameter matrices are solved.Then,the distortion coefficient is estimated according to the relationship between the real image and the detected image coordinate with initial camera parameter fixed.At last,LM algorithm is applied to the global optimization and the more accurate intrinsic and extrinsic parameters are obtained.Experimental results indicate that error of equivalent focal length is less than 0.1% and error of principal point is less than 0.5 pixel when Gaussian noise with less than 0.5 pixel standard deviation is added to the simulation image data.Under the same Gaussian noise level the more the pictures used,the smaller the standard deviation of camera parameters.Compared with other calibration methods 3D calibration board proposed in this paper is more accurate.More stable camera parameters can be obtained by using more pictures of the calibration board.