基于聚类分析的管路图像亚像素边缘提取算法

To promise stress-free and precise assembly for bend tubes, the geometric parameters shoule be measured during the manufacturing stage. Machine vision based measurement for bend tube is a rapid and accurate method, which is widely used in the field of 3D measurement. Aiming at the problem that the traditional extraction in this method could not obtain pipeline edge accurately, a sub-pixel edge extraction method for tube's image even was proposed under a complex illumination condition. The extraction was explained in four steps: the low-frequency noises were filtered with spectral filtering method, and a clustering analysis method was applied to segment the tube's region precisely; the pixel edge was obtained with image morphology method. The surface fitting method was employed to fitting the variation of local gray values to realize the sub-pixel edge points. Experiments results showed that the method could extract sub-pixel edge accurately and reliably. The accuracy reached 0.04 pixel width, which could provide the accurate edge for reconstructing tube's 3D model.