基于方向梯度直方图的螺纹连接件端面缺陷检测方法

Detection method of end face defects of threaded connectors based on histogram of oriented gradient

针对螺纹连接件端面表面的反光性、阈值选取的不确定性,提出了一套图像处理方法.使用中值滤波和图像锐化增强等方法处理图片,解决背景与前景差距较小的问题;选取利用方向梯度直方图确定图像上的边缘方向,实现对目标的直线轮廓提取,与已知目标的轮廓模型结合实现对非直线目标轮廓的提取.该处理方法可以自动提取目标的轮廓,而不需要设定阈值,且对亮度、对比度和噪声变化的适应性较好.对2 500个报警器底盘做的检测实验结果表明,采用该方法能去除噪声,与Canny轮廓提取法相比,能更准确且稳定地提取出螺纹连接件端面的边缘特征;对螺纹连接件端面磨损和歪斜的检测准确率分别达到了99.4%和98.6%,满足工业的准确率和实时性的需求;同时可减少人工成本,提高检测速度.

To tackle the issues of the surface reflection of threaded connectors and the uncertainty of threshold selection, a set of image processing methods are proposed, whereby median filtering and image sharpening are used to process the image to solve the problem of small gap between background and foreground. The histogram of oriented gradient is applied to determine the direction of the upper edge of the image to extract the linear contour of the target, and the contour model of the known target is combined to extract the contour of the non-linear target. This method can automatically extract the contour of the target without setting a threshold, and has good adaptability to the changes of brightness, contrast and noise. In this paper, 2 500 alarm chassis detection experiments are carried out. The experimental results show that this method can remove the noise and extract the edge features of threaded connector surfaces more accurately and more stably than canny contour extraction. The accuracies of the proposed method for threaded connector’s end face wear and skew detection are 99.4% and 98.6% respectively, which meet the requirements of real-time industrial accuracy. In addition, it can reduce the labor cost and improve the detection speed.