基于边缘方向投影的棋盘格角点亚像素细化

Checkerboard Corners Sub-Pixel Refinement Based on Edge Direction Projection

随着视觉测量技术在工程中的推广,越来越多的视觉标定和测量需在车间现场由非专业人员执行,这会造成拍摄的棋盘格图像中包含较多的噪声。为了实现噪声下棋盘格角点稳健、精确的亚像素细化,提出一种基于边缘方向投影的棋盘格角点亚像素细化方法。首先基于非极大值抑制算法计算初始边缘方向,然后基于最小加权二乘拟合法细化边缘方向,最后基于边缘方向最大投影细化棋盘格角点的亚像素坐标。结果表明:在高质量的棋盘格图像中,所提方法的棋盘格边长测量偏差的最大值均小于0.021 mm,棋盘格边长测量偏差的均值均小于0.006 mm;在高斯噪声和角点污染的棋盘格图像中,所提方法的棋盘格边长测量偏差的最大值均小于0.05 mm,棋盘格边长测量偏差的均值均小于0.02 mm。

With the popularization of visual measurement technology in engineering,more and more visual calibration and measurement need to be carried out by non-professionals in the workshop site,which will cause checkerboard images to contain more noise.In order to achieve robust and accurate sub-pixel refinement of checkerboard corner under noise,a checkerboard corners sub-pixel refinement method based on edge direction projection is proposed.First,the initial edge direction is calculated based on the non-maximum suppression algorithm.Then,the edge direction is refined based on the least weighted square fitting method.Finally,the sub-pixel coordinates of the checkerboard corners are refined based on the maximum projection of the edge direction.The results show that in the high-quality checkerboard images,the maximum measurement errors of checkerboard edge length are all less than 0.021 mm,and the average measurement errors of checkerboard edge length are all less than 0.006 mm.In the checkerboard image with Gaussian noise and corner pollution,the maximum deviation of checkerboard edge length measurement of the proposed method is less than 0.05 mm,and the average deviation of checkerboard edge length measurement is less than 0.02 mm.