Extraction of shape and metric information from image data
Abstract
This thesis research involves issues of extracting shape and metric information from image data. There are three topics which are approached from different aspects. A shape classification system has been developed using the Fourier descriptors of a closed boundary and generic shape features. It can trace using chain code, crack code and polygonal fitting as a user's option, from which Fourier descriptors and generic shape features are extracted. Unknown objects can be identified by library matching of the Fourier descriptors. Generic shapes are independent of any prior knowledge and confidence number ranging from 0 to 100 are assigned for each generic feature: similarity to circle, ellipse, rectangle, and square; bilateral symmetry, axial symmetry, 3-fold and 6-fold symmetry. These confidence numbers are accessible by a rule-based system to interpret the image data and a prototype classification system is demonstrated using generic shape features. A second consideration is that shapes may be distorted due to the nonlinear transformation of the imaging process. Under perspective transformation, a circle on a planar object appears as a conic (ellipse, hyperbola, parabola) in the camera image, and the coefficients of the conic equation vary as the plane orientation changes. Theoretical analysis of the relationship between these coefficients and the plane orientation is made and the application for pose determination of a robot using a standard mark is addressed. The final issue is the determination of plane orientation from the analysis of the image of a projected circular pattern. By evaluating the coefficients of the imaged ellipse resulting from the projected circle, plane orientation can be determined at high speed with reasonable accuracy. The technique can be employed to inspection or automatically positioning robots relative to surfaces. A Gaussian distributed pattern can also be employed as a projector or laser output, and the ellipse is constructed by thresholding the image at a certain level. Conic fitting is made using subpixel accuracy to estimate surface orientation.
Degree
Ph.D.
Advisors
Mitchell, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.