基于样条变换多元回归模型的摄像机标定算法

Camera calibration algorithm with multiple regression model based on spline transformation

针对摄像机隐式标定算法中存在的样本点数少和精度要求高的矛盾,利用样条变换多元回归模型,提出一种改进的摄像机隐式标定算法。首先采集标定图片中有限个标定样本点的像素坐标和相应的世界坐标,然后通过样条变换建立像素坐标和世界坐标间的多元回归模型,获得二维图像在世界坐标系中的三维信息,完成摄像机的标定。算法发挥上述模型在标定样本点有限的条件下,对非线性关系映射能力强、局部性质独立的特点,实现整体标定效果的改善。理论分析和实验结果均表明,与基于多项式拟合的标定算法相比,该算法的标定速度与精度显著提高。该算法已成功应用于移动机器人视觉导航系统,提高了导航信息计算的准确度。

A modified camera calibration algorithm is proposed using the multiple regression model based on spline transformation to solve the contradiction between finite samples and high precision.By collecting the pixel coordinates and their corresponding world coordinates of limited sample points,the multiple regression model based on spline transformation from pixel coordinates to world coordinates is established;then the three-dimensional information of a two-dimensional image in the world coordinate system is obtained and camera calibration is implemented.Under the conditions of finite samples,the algorithm takes advantage of the strong non-linear mapping ability and independent local characteristic of the multiple regression model to improve the effects of camera calibration.Theoretical analysis and experimental results show that the proposed algorithm has faster speed and higher precision compared with the algorithm based on polynomial fitting.The algorithm has been applied to the vision navigation system of mobile robots and successfully improves the accuracy of navigation information calculation.