基于直线的几何不变性标定摄像机参数

计算机视觉通常采用针孔摄像机模型，但对于存在较大畸变的鱼眼镜头或广角镜头来说，会造成图像中同时存在透视变形和像差畸变。解决此问题的传统方法一般是采用标准网格板来标定摄像机参数，但需要较多的已知信息。为了进行精确的标定，提出了一种新的标定方法，该新方法不需要任何空间3维信息，即可用单幅普通图像来标定摄像机的像差系数及内参数，并可将畸变图像校正到相似变换。为了纠正像差畸变和计算消影点，该方法采用了直线的射影不变性，即共线点的投影仍然共线，平行直线束的投影相交于一点的性质；为了纠正透视变形，还采用了直线的相似不变性，即正交直线的夹角在相似变换中仍然保持正交的性质。用该方法标定的摄像机的参数包括像差系数、焦距、主点和纵横比，同时将图像纠正到了相似变换。用实验室图像和室外图像进行了仿真实验都得到了精确、可靠的结果。

A Method of Camera Calibration Using Geometrical Invariability of Line

Many algorithms in computer vision usually use a pin-hole camera model. An image is caused by both projective transform and the lens distortion as fish-eye lens or wide-angle lens have large distortion. Unlike the traditional methods which need much information of the scene,a novel approach using geometrical invariability of line to perform camera self-calibration from a single image is presented. In order to correct lens distortion and calculate vanishing point from the image, projective invariants are used: collinear points should be in the same line, Parallel line should be intersect at one point; metric invariant is used: the included angle of orthogonal straight lines should be rectangular. So camera lens distortion and internal parameters are calibrated. The result of emulation and real images are accurate and reliable with this method.