I am trying to develop an application that will give me the orientation and offset of an integrated circuit (IC, chip) compared to a reference image. I can be sure that the device is within its width and about 30 degrees of the target.
Extracting keypoints then using FlannBasedMatcher works well if it's an identical device (same markings, lighting etc) but offering up a different item.... oh dear. Because the device is so uniform in structure, each pin is identical to the next, but surface markings are different, matches are almost at random. The surf feature detector works but the features it detects are too small and numerous and, again, there's no uniformity to the matching. I can't work out how to make it look for larger features. I've tried Sobel differentiation and produced some very good outline images then used a harris corner detect but I still can't work out how to match like for like on a structured basis i.e. match pin 1 to pin 1 and so forth.
Does anyone have any ideas of what sequence of steps I should take.