An automated system drills the outer moldline holes on a military aircraft wing. Currently, the operator manually checks countersink diameter every ten holes as a process quality check. The manual method of countersink inspection (using a countersink gauge with a dial readout) is prone to errors both in measurement and transcription, and is time consuming since the operator must stop the automated equipment before measuring the hole.
Machine vision provides a fast, non-contact method for measuring countersink diameter, however, data from machine vision systems is frequently corrupted by non-gaussian noise which causes traditional model fitting methods, such as least squares, to fail miserably. We present a solution for circle measurement using a statistically robust fitting technique that does an exceptional job of identifying the countersink even in the presence of large amounts of structured and non-structured noise such as tear-out, scratches, surface defects, salt-and-pepper, etc. The method is based on an easy to implement iterative algorithm using edge detection. Convergence takes less than one second. Using this technique, we have been able to repeatably identify countersink diameters to better than 0.1 pixels of the image (< 0.02mm for a 30mm field of view) in factory environments.