针对Harris算子在提取标定图像中的特征点时可能出现的伪角点, 提出了一种新的自动修正算法.通过对伪角点邻域内图像进行X和Y两个方向上的一维亚像素扫描, 然后用基于正交距离的直线或曲线拟合算法计算两条边缘线条, 最后用两边缘线条的交点来估计正确角点位置.本文对机器视觉标定中的棋盘格图像进行大量实验, 结果证明该方法具有较好的鲁棒性, 达到了预期的两个目的：1)修正图像中的伪角点; 2)得到高准确度的亚像素角点.

A novel method is depicted in this paper to handle the probably existed false corners from the feature extraction on calibration image using Harris operator. It performs x and y directional sub-pixel scans in the neighborhood of false corners respectively, then line fitting or curve fitting method based on orthogonal distance is employed to compute the intersection of two edge lines for the purpose of replacing the location of false corner. A lot of experiments on the images of chessboard for machine calibration show that the method is of robustness, and achieve two expectant objectives: 1) amend false corners in result; 2) get the sub-pixel level corners of high accuracy.