Accurate Edge Localization of Complex Workpiece Based on Coarse-to-Fine Principle

The accurate measurement of the heterogeneous surfaces of the complex workpiece is very important. Therefore, it has become a hotspot issue to measure the dimensions of the complex workpiece with high efficiency and accuracy nowadays. Taking the piston as an example, we measure the dimensions of several surfaces by using the approaches of digital image processing. The key point of the measurement is to determine the precise edge of these surfaces. However, in general, it is not an easy job to obtain the accurate edge due to the reflected light of the surface. Otsu binarization is a popular method in computing the surface edge. However, the Otsu binarization does not function well in many cases. After carefully analyzing the advantages and disadvantages of the Otsu method, we put forward a novel technique to obtain the precise edge by employing the coarse-to-fine principle. Firstly, we preprocess the image; including converting the image to a grayscale image and stretching the image gray level. Secondly, we use the sub-pixel interpolation to magnify the processed image by 1.3 times. Thirdly, we employ the Otsu method to perform the binarization operation on the magnified image, perform horizontal histogram projection and vertical histogram projection, respectively. Fourthly, estimating the radius length and the circle central coordinates. Fifthly, we estimate peripheral points of the top circle according to the radius length and the circle central coordinates, and take this point as the center and get the appropriate proportion rectangle according to the radius length. Finally, we intercept the image in the rectangle and perform the local binarization operation. The method fully takes into account the situation that we cannot get the accurate edge through global binarization for severe reflection of the workpiece which causes the uneven gray-level distribution. Our method improves the detection accuracy of a single pixel by 0.036 millimeters when the image resolution is not high.