VISION FOR INTELLIGENT INSPECTION AND ASSEMBLY: SOFTWARE TOOLS AND THEORY (IMAGE-PROCESSING, CONTOUR-TRACING, FLEXIBLE AUTOMATION)
Abstract
This thesis includes topics from three different strata of image processing. In the area of feature extraction, a new form of contour description called crack code is presented. It is an eight-direction chain code like that of Freeman, but has a shorter diagonal move providing for more accurate representation of perimeter and area directly. Crack code by definition cannot trace back upon itself providing a deterministic process which has a nearly parallel implementation. Insight is gained into the Freeman chain code representation through the development of conversion algorithms between chain and crack code. Specific forms for perimeter correction and measurement tolerance are developed and tested for crack code descriptions of straight line boundary sections. The area of automated inspection is represented by the application of the threshold decomposition technique to crack code contour description. In this way, the accuracy of silhouette perimeter and area measurements can be increased by a factor dependent on the number of intensity quantization levels producing sub-pixel results. Developments for straight line length and position estimation are included for images with a grey level continuum along with considerations for images having a finite number of quantization levels. It is shown that the perimeter error is independent of boundary position and the boundary position location error is 0 for the continuous case. Both results become unbiased estimates in the presence of grey level quantization and tolerances on the measurements are shown both theoretically and experimentally. In the area of image processing system design, consideration is given to the applications of intelligent simulation in flexible automation work cell design. It is presently common to design a product for ease of manufacture, and that idea is extended here to include the design of manufacturing processes for ease of automated visual inspection. An object-oriented, user-friendly programming environment is developed which allows the symbolic simulation and manipulation of tools, products, and cameras through three dimensional wire frame graphic models. The same environment can then be used to monitor the process in the work cell developed while in progress to permit expert recovery from unexpected situations.
Degree
Ph.D.
Subject Area
Electrical engineering|Artificial intelligence
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.