Date of Award

1988

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

First Advisor

Edward T. Lee

Abstract

This thesis addresses the problem of recognizing partially occluded two dimensional objects. The goal is to develop a system which is able to identify and locate several overlapping objects in the scene. To achieve this goal, the system must perform the following specific tasks: (1) storing useful information about objects in some format, which is often referred to as the process of object representation or model formation (2) matching procedure based on the object representation, and (3) efficient search of the best matching. This thesis presents a new approach to accomplish these tasks. Polygonal approximation is used to represent an object in this research. The accumulated lengths of line segments, s, and the accumulated sizes of turning angles, $\theta$, along the boundary from some starting point are extracted. The boundary of an object is then described as an equation $\theta$ = f(s). As algorithm shows, matching objects under s-$\theta$ space will be simple and effective. To avoid exhaustive matching in the recognition process, index diagrams of the features characterizing the boundary are established. Once the features of some unknown object are detected, the possible objects which might produce the best matching can be efficiently retrieved from this scheme.

Pages

132

Share

COinS