SMT元件影像检测关键算法

提出了一种基于视觉的贴片元件几何特征参数的检测方法。该方法首先采用最大外接矩形的方法实现元件的粗定位及确定边缘的分割点。接着采用Canny算子和Zernike矩边缘检测算子实现边缘的亚像素精确定位。然后，利用分割点将边缘分割成4部分，分别进行直线拟合，得到边缘的直线方程，利用直线和圆弧相切的关系，求得直线和圆弧的切点，采用最小二乘法对两切点间的边缘点进行圆弧拟合得到圆弧的精确值。同时，采用快速傅立叶变换，利用变换后图像的特征，实现端面图像中条纹方向的判定。实验中测得亚像素边缘点的定位精度为0.03pixel，直线拟合精度为0.03pixel，圆弧拟合精度为0.05pixel，端面条纹判断的准确率为100%。理论分析和实验结果表明：本文提出的最大外接矩形分割法、亚像素定位法、直线圆弧拟合法及条纹方向判断法能很好的满足贴片元件几何特征参数自动视觉检测的稳定可靠、精度高及实时性的要求。

Key Algorithms of Image Inspection for SMT Components

A visual method to calculate all geometrical parameters of chip components is proposed. The orientation obtaining of the chip component and the edge point sorting are realized by calculating the maximum external rectangle. Then the edge points are detected by Canny operator and relocated by Zernike moments operator to achieved subpixel location precision. The edge points are sorted into 4 parts by sorting points and each of which is fitted into a line. Tangent points are calculated by using the tangent relations of two lines and the arc. Edge points between two adjacent tangent points are fitted into an arc by least squares arc fitting method and as a result the arc precise parameters are obtained. Meanwhile the transverse image of chip component is transformed by using Fast Fourier Transform (FFT) and according to the characteristics of transformed image the stripe direction can be judged correctly. The experimental results demonstrate that: the precision of subpixel location is 0.03pixel, linear fitting precision is 0.03pixel, and the arc fitting precision is 0.05pixel. The judge accuracy of the stripe direction of the transverse image is 100 percent. Methods and algorithms adopted in the paper can meet satisfy the measure requirements of stabilization, higher precision and real time for inspecting the chip component geometrical parameters.