基于单位视球的鱼眼相机标定方法

针对鱼眼镜头的大范围视场,提出一种新的鱼眼相机标定方法.利用2组相互垂直的平行线来进行鱼眼相机标定.依据相机视场和成像边缘得到内部参数的估计值后,将图像平面上提取的角点反投影到单位视球（viewing sphere）上,2组平行线在单位球面上的2种几何特性提供外部参数估计的解析解.利用角点在图像平面的重投影误差来,得到优化后的所有参数.对3种常见的鱼眼投影模型分别进行关于标定图像数量和噪声水平的仿真实验.从仿真结果来看,在合理的噪声范围内,当用于标定的图像大于5张时,可以得到精度较高的标定结果.利用视场角185°的鱼眼相机来进行标定,并进一步的利用已知结构的立方体模板来验证外参估计方法.与加州理工学院提供的标定工具箱相比较，结果表明该算法不确定度较低,提供更为准确的标定结果.

Calibration of fisheye cameras based on the viewing sphere

With respect to the large field of view (FOV) of a fisheye camera, a new calibration method was proposed, This method takes advantage of two mutually orthogonal sets of parallel lines. The intrinsic parameters could be initialized by the field of view (FOV) and imaging boundary, and corners extracted in the image plane could be back-projected to a viewing sphere. Two geometric properties of the two line sets on the viewing sphere provided a closed-form solution for extrinsic parameters. Finally, all the parameters were optimized according to the re-projection errors of the corners. Three commonly used fisheye projection models are simulated with respect to the image number and noise level. The simulation shows that, within meaningful noise, the method can achieve high accuracy when the image number is larger than 5. A fisheye camera with 185°FOV is also calibrated, and a geometry-known trihedral object is further tested to validate the estimation of extrinsic parameters. The experiments compared to the popular fisheye calibration toolbox from Caltech prove that with a lower uncertainty, the proposed method is more accurate than Caltech’s toolbox.