Our method assumes that the images has been normalized and binarized. Normalization, in this context, is geometric correction of an image. Its purpose is twofold:
The normalization is done by mapping onto the image corners the centroids of four circular markers located close to the corners. (See figure 3.) This procedure and the subsequent binarization (thresholding) are presented in [6]. The measurements are done with subpixel accuracy, as described in the same paper. Examples of measurements as well as the dimensions measured in case of E-cores are given in section 4.2.
Note that the our original version of the matching algorithm presented in [6] has a limited scope. Since this algorithm was created for specific products, E-cores, extending this algorithm to other shapes requires complete re-writing of the system. In addition, the original matching algorithm [6] cannot be extended to 3D. The new general matching methodology presented here is free of these deficiencies.
The main steps of the new matching method are as follows.
Step 6 resolves ambiguities arising from the integer approximation of the distance. Also, it discards possible false minima when symmetric contours are matched. For such contours, many points may coincide in a wrong orientation because of symmetry, while the rest of the points may not coincide at all.
Alternatively to the criterion of step 6, the pose with the largest number of measured points lying within the minimum median HD can be used.
Figures 6 and 7 illustrate the operation of the matching algorithm. Step 1 is exemplified in figure 6, where the contour of a reference ferrite core is shown along with its distance transform. Figure 7 explains the matching of a measured contour against the reference contour of figure 6. In this figure, the measured contour is shown in one of the relative positions, overlaid on the DT of the reference contour. (Step 4.)
![]() |