使用12像素对称模板的棋盘格内角点检测

棋盘图像的角点提取问题往往决定着三维测量中摄像机标定的精度。针对SUSAN（Smallest Univalue Segment Assimilating Nucleus）算法无法区分棋盘标定板内角点与边缘点的缺陷，提出一种12像素对称灰度模板检测算法。该算法首先根据棋盘格内角点周围像素的中心对称性分布，设计一种12像素对称USAN模板，可以迅速区分出内角点与边缘点，同时将内角点与平坦区域作为候选点。再结合灰度均方差算子，利用平坦区域灰度方差较小的特点将其剔除，最终实现对棋盘格内角点的高效检测。同时，该算法在检测过程中完全摒除易受外界因素影响的外圈角点，以保证角点提取时的精度。实验结果表明:新算法对9阶棋盘格的检测时间为1.244577s；用于张正友标定方法之后，得到的检测重投影误差仅为[0.3，0.3]像素。这两项指标，均优于传统SUSAN算法。

Internal corner detection of chessboard image for camera calibration based on 12 pixels symmetrical template

The problem of chessboard image corner extraction always determined the three-dimensional measurement's accuracy of the camera calibration. By analyzing the defect for SUSAN (Smallest Univalue Segment Assimilating Nucleus) algorithm that could not effectively distinguish the chessboard internal corners and edge points, the authors made use of the symmetry of the pixels around the internal corners, and proposed a symmetrical 12 pixels gray template detection algorithm. Firstly, a symmetrical 12 pixels USAN template was designed for fast distinguishing the internal corners and edge points. Meanwhile, both of the chessboard internal corners and smooth region would be treated as the candidates. Then the less gray variance of smooth region could be used to abandon them. At the same time, the proposed algorithm abandoned the external corners of the chessboard, which were very sensitive to the external factors, ensuring the precision of the corner extraction process. Experimental results show that the new method detects the nine order chessboard image by 1.244 577s, and its reprojection error was just [0.3, 0.3] pixels in Zhang's camera calibration. Both of these two indicators are better than the traditional SUSAN algorithm.