中文摘要：

提出一种基于灰度差异的棋盘格角点自动检测方法，该方法利用一种圆形角点检测器，在不对棋盘格图像做预处理情况下，能够准确检测有一定噪声或者较模糊的图像上的棋盘格角点。首先，根据角点及领域附近点的灰度差异，采用角点检测器初步检测角点；其次，根据棋盘格角点张角近似为直角和角点灰度对称性，对初步检测的角点进一步筛选获得棋盘格目标检测角点；最后，采用灰度平方重心法将像素级棋盘格目标检测角点亚像素级化，精确确定角点位置。实验结果表明，该方法在图像较模糊的情况下能够较精确检测出角点，角点重投影精度达到0．1像素级，并能获得形成角点的张角角度值；该方法可以应用于具有特殊角度的角点检测，且由于运算速度快，误差小，能够应用于机器视觉摄像机的在线标定。

英文摘要：

A checkerboard pattern corners auto-detection method so-called circle corner detector for camera calibration is presented based on the grey-level difference, and it can detect the checkerboard corners from the noise-image or blurred image without any image preprocessing. Firstly, a circle cotner detector is designed for checkerboard corners detection, getting a preliminary resu the grey level difference between the corner and its neighboring points. Secondly, the pixel level corners can be obtained through sifting the preliminary results according to the feature of the checkerboard pattern corners whose angle are nearly right-angle and grey value nearby is symmetry. Finally, a sub-pixel level corner can be achieved by grey-level squared weight center method and therefore correct position of the checkerboard corner can be determined. The experiment results show that a high accuracy of checkerboard pattern corner detection has been achieved, and the corner reprojection error is less than 0.1 pixel level even though the image is blurred. For the circle-corner-detector based checkerboard pattern corners auto-detection method can obtain the splayed angle value of the corner, it can also be applied for various corner detection with special angles, and also for its high calculating speed and high accuracy, it can be appropriate for on-line camera calibration in machine vision.