基于改进棋盘的角点自动检测与排序

考虑采用传统黑白棋盘进行相机自动标定时角点排序结果受标定模板旋转角度影响较大,本文设计了一种改进的棋盘标定模板及相应的角点自动检测与排序算法。新的标定模板通过增加4个长方形边界滤除复杂背景,增加沙漏状图形确定排序原点,从而使得排序结果适应于模板的旋转。针对新的标定模板提出了一种基于对称象限灰度交叉熵的角点检测算法,该算法通过抑制局部非极大值以及筛选矩形边界实现了角点的像素级定位,然后采用Frostner算子解算了角点的亚像素坐标。针对角点检测结果,采用曲线拟合并结合角点至原点的距离信息实现了角点自动排序。实验结果表明:得到的角点检测结果正确,亚像素解算坐标与Matlab标定工具箱的检测结果误差小于0.8pixel,排序结果对标定模板的旋转具有不变性,易于实现在线标定。

Automatic detection and sorting of corners by improved chessboard pattern

In automatic camera calibration with traditional black and white chessboard patterns, the corner sorting results are usually influenced by the rotation angle of the calibration pattern. Therefore, this paper designs an improved chessboard pattern and corresponding automatic corner detection and sorting algorithm. In the new pattern, four rectangular boundaries were added to filter the complex background, and a double-triangle mark was used to determine the original point of the corners so as to adapt to the rotation of the pattern. A corner detection algorithm based on cross entropy of the symmetrical quadrant was proposed to implement the corner position with the accuracy of pixel level by local non-maximum suppression and rectangular selection. Then, the Frostner operator was used to calculate the sub-pixel coordinates of the corners. According to the detected corner, the curve fitting method was employed to realize the automatic corner sorting with the distance information between the corners and the origin. Experiment results show that the corner detection results are correct and the sub-pixel coordinate error between the new method and Matlab Calibration Toolbox is less than 0.8 pixel unit. Moreover, the sorting results show an invariance to the pattern rotation, which verifies that the method is suitable for online auto camera calibration.