为了进一步提高相机参数精度, 简化标定过程, 在张正友平面标定算法的基础上, 提出了一种较简单的5点标定算法。本文利用一个矩形作为标定模板, 将其4个角点和中心点作为待匹配点。仅利用不同角度拍摄的10(10~20幅均可)幅图像, 即可求解全部相机内外参数。首先, 为了减小数据对结果的影响, 对获取的5个点的图像坐标进行归一化, 将点坐标控制在以原点为中心, 2为半径的圆内; 其次, 相机成像模型未考虑畸变, 考虑到单应矩阵可能会有零元素, 故按照9个自由度来求解; 再次, 求解出相机的内外参数, 将求得结果的平均值作为非零因子λ的值; 最后利用Levenberg-Marquarat(LM)算法对所有内外参数进行整体优化。选用不同噪声水平的模拟实验和真实图像进行测试, 结果显示该算法比张正友棋盘格算法的精度要高。

A simpler 5-point calibration algorithm was proposed to further improve the accuracy of camera parameter and simplify the calibration process, on the basis of Zhang Zhengyous plane calibration algorithm. In this method, a rectangle was used to serve as a calibration template and the four angular points and central point were selected as the points to be matched. All internal and external parameters could be solved with only 10 (10-20 all were ok) images shot in different angles. First, to decrease the influence of data on result, normalization shall be carried out for image coordinates of five points obtained, where the coordinate of point shall be controlled within the circle with the origin as the center and 2 as the radius; then, because it didnt consider distortion in imaging model of camera but considered that there might have been zero element in homography, caculation was solved in accordance with 9 degrees of freedom; again, the internal and external parameters of camera were caculated and the value of non-zero divisor λ was the mean of result acquired; finally, global optimization for internal and external parameters was carried out with Levenberg-Marquarat (LM) algorithm. Through simulation test with different levels of noise and test of true image, the accuracy of this algorithm is higher than Zhang Zhengyous checkerboard algorithm.