Date of Award

1999

Degree Type

Dissertation

Degree Name

Ph.D.

Department

Electrical and Computer Engineering

First Advisor

Jullien, G. A.

Keywords

Engineering, Electronics and Electrical.

Rights

CC BY-NC-ND 4.0

Abstract

One of the aims of industrial machine vision is to develop computer and electronic systems to replace human vision in quality control of industrial production. Traditionally these systems consist of a line scan camera, host computer, frame grabber and one or more dedicated processing boards. The work reported in this thesis develops defect detection algorithms for real-time processing of the camera video stream. The processing system is mounted inside the camera and provides sufficient defect detection capabilities to eliminate the need for an external frame grabber and other associated host computer peripheral systems. The system is targeted for web inspection but has the potential for broader application areas. The output data from the camera is reduced by many orders of magnitude by only transmitting the "interesting" pixels of the image to be processed, and this can significantly reduce both the downstream processing hardware required and the bandwidth of the digital data received from the camera. The use of such special purpose cameras has the potential not only to improve the performance of machine vision systems for a wide variety of applications, but to improve the economic viability of these applications through reductions in hardware cost and complexity. This real-time system must perform all of the required operations at the video bandwidth of the camera, and the work reported in this thesis uses hardware associated with the in-camera processing system, developed in the VLSI Laboratory at the University of Windsor, which includes programmable logic (Field Programmable Gate Array) directly connected to the video stream, and ancillary signal processing and control hardware (a DSP chip). These hardware limitations apply constraints to the algorithms, and we are almost always unable to use traditional image processing algorithms; rather we choose and develop algorithms based on their potential for identification based on minimal storage of a pixel-serial raster data. In this thesis we report the following novel developments: (1) For non-textured background materials, three algorithms have been developed for the in-camera system: two (or multi) level thresholding; zero order background tracking; and delta modulation background tracking. (2) Auto-regressive techniques have been developed and implemented as a statistical approach to analyze textured backgrounds and to identify possible defects. This method of analysis has been extensively used to study visual textures. In the simplest form, the image is scanned to provide a one dimensional series of gray level fluctuations, which is treated as a one-dimensional stochastic process evolving in "time". In a more comprehensive form, a pixel value is assumed to depend upon a certain part of its neighborhood. The coefficients of dependence are extracted using time series analysis techniques. (3) A novel algorithm for defect detect detection based on fuzzy fusion of texture features is developed, simulated and successfully implemented on the experimental test setup. Conventional approaches for web defect detection involve making "crisp" decisions for image analysis and recognition where imprecise or incomplete specifications are usually either ignored or discarded. The fuzzy logic algorithm uses imprecise or ambiguous image data caused by instrumental error or environmental noise such as dust or small variations in illumination to obtain a precise result. The developed algorithm can be applied to both textured and non textured materials and offers superior performance over traditional template matching methods.Dept. of Electrical and Computer Engineering. Paper copy at Leddy Library: Theses & Major Papers - Basement, West Bldg. / Call Number: Thesis1999 .H35. Source: Dissertation Abstracts International, Volume: 62-10, Section: B, page: 4686. Adviser: G. A. Jullien. Thesis (Ph.D.)--University of Windsor (Canada), 1999.

Share

COinS