Efficient and accurate 3-D object representation and recognition method based on CAD

Abstract

In this thesis, we automatically build the efficient and accurate 3-D object model and the matching strategy based on the CAD. At first, we represent the object with a set of continuous surfaces using the extended boundary model which efficiently represents the geometric and topological information of curved objects. Then, we represent the view space of each object using the nonuniform view sphere tessellation method with a set of view-points. And we extract the viewpoint dependent features of surfaces and surface discontinuities by analyzing the CAD model from each viewpoint. To accurately and reliably extract the viewpoint dependent features, we propose the object precision method. The accuracy of the proposed object precision method is same with the CAD model and is independent of the viewpoint. Objects are represented with a set of identity indexing trees and viewpoint indexing trees. Identity indexing tree is to hypothesize the object by hashing the objects with the index values of the features. Thus, it uses the identity sensitive features. Viewpoint indexing tree is to locate the object by hashing the viewpoint with the viewpoint sensitive features. Objects are recognized with the geometric hashing technique whose efficiency is determined by the utility and matching order of each feature, the verification order of the hypothesis and the termination condition of the verification. For this work, we propose the formal definition of the measures for feature utility and hypothesis certainty. Identity-priority and viewpoint-priority are used for the measures of feature utility. Identity-priority - defined by the identity-sensitivity and the reliability - is used to rank the extracted features for hypothesizing the identity and viewpoint-priority - defined by the viewpoint sensitivity and the reliability - is used to rank the extracted features for locating the object. Identity-verifiability and viewpoint-verifiability are used for the measures of hypothe...