MODEL-BASED 3-D VISION
KANG SING BING
KANG SING BING
Citations
Altmetric:
Alternative Title
Abstract
This thesis is divided into two pans. The first part discusses in length the state-of-an techniques in 3-D vision, specifically 3-D data acquisition, representation, and recognition. The second part, which forms the basis for this thesis, deals with a specific method of identifying 3-D objects based on a single view. A new approach of recognizing 3-D objects based on fuzzy subset theory is proposed in this thesis. This method uses Zernike moment invariants of the silhouette of the unknown object to form a set of fuzzy-weighted quantities called fuzzy quaternions. The database of the models is generated by first creating the octree representation of the 3-D objects based on three orthogonal views of the objects and employing the volume intersection method. Subsequently, by using the multi-level boundary search algorithm, the object of interest can be displayed quickly at a given arbitrary viewpoint. The database is constructed by calculating the fuzzy quaternions of each object at 120 discrete and suitably spaced viewpoints. In the calculation of the Zernike moment invariants which make up the fuzzy quaternions, Green's theorem in the plane is used to speed the computation process. The accuracy of tile resulting digitized contour integral is enhanced by applying the p-correction factor (from the average error of about 3.1 % to about 0.7%). Finally, in the recognition process, the set of quaternions is calculated for the silhouette of the unknown object. This is then matched against those in the database. To speed up the recognition process, a modified version of the Nelder-Mead simplex algorithm is utilized to minimize the measure of discrimination called the dissemblance index. Results have indicated the success of this method, with the average success rate of about 95% (when quaternions of up to the third order are used).
Keywords
Source Title
Publisher
Series/Report No.
Collections
Rights
Date
1990
DOI
Type
Thesis