Recognition of Parts and Their Orientation for Automatic Machining, Handling, and Inspection

摘要

This report describes a technique whereby a flexible robot with vision can recognize a workpiece in various orientations and under certain background conditions of overlap and clutter. A registration technique developed previously to match aerial imagery to maps was tested for matching parts imagery to the blueprints. Primitive features are automatically extracted from the images and a set of vectors derived which represents real or abstract edges in the image. Each vector can be used to determine a unique rotation and translation mapping the image edge onto a model edge. A cluster formed in the space of all such mappings reveals a mapping that matches many edges of the part onto many edges of the model and thus the aspect of the imaged part is known. This registration technique was tested on images of carburetor covers and hinges. In general the results were satisfactory with all types of imagery tested. In all but one case at least one object in the image was correctly acquired.