Abstract: In view of current edge detection algorithms have deficiencies, such as low positioning accuracy, slow processing speed and poor antinoise performance, etc, this paper presents a method of axial parts size detection based on image edge high accuracy positioning. By using the improved self-adapting median filtering algorithm, the improved Kirsch operator, and the fittiing method that uses quadratic function to approach the Gaussian curve in the grayscale gradient direction of image edge, this method realizes the image edge subpixel high accuracy positioning and improves the precision of dimensional measurement. Through the application of computer vision inspection of valve size, it proves that the accuracy of this algorithm presented here is precise and stable, meeting the requirements of high precision visual detection.

Key words: median filter, curve fitting, sub-pixel, computer vision, size measurement

摘要： 针对目前的边缘检测算法存在定位精度低、处理速度慢、抗噪性能差等缺陷，提出了一种轴类零件尺寸检测的图像边缘高精度定位方法。该方法采用改进的自适应中值滤波算法、改进的Kirsch算子和在图像边缘灰度梯度方向上进行二次函数逼近高斯曲线拟合方法，实现了图像边缘亚像素高精度定位，提高了尺寸检测精度。通过对气门尺寸的计算机视觉检测实际应用，证明提出的算法精确且稳定，满足高精度视觉检测要求。

关键词: 中值滤波, 曲线拟合, 亚像素, 计算机视觉, 尺寸检测