Segmentation and classification of THCs on PCBAs

Abstract

The dramatic increase of electronic waste requires automatic recycling, including technologies from machine vision. A framework for segmentation and classification of THC (through-hole components) mounted on PCBAs is presented, using both RGB and depth frames from the Kinect sensor by Microsoft. A segmentation approach, combining local and global features in a flexible manner, is shown to optimize a freely definable cost function globally. We interleave segmentation and classification as we form the final components using a simple, yet robust shape model.