Full text
| |
| Author(s): |
Mercedes Rocio Gonzales Marquez
Total Authors: 1
|
| Document type: | Doctoral Thesis |
| Institution: | Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Elétrica e de Computação |
| Defense date: | 2009-00-00 |
| Examining board members: |
Olga Regina Pereira Bellon;
Antonio Castelo Filho;
Alice Maria Bastos Hubinger Tokarnia;
Clésio Luis Tozzi
|
| Advisor: | Wu Shin Ting |
| Abstract | |
The crude registration of two partially overlapping range images from rigid objects consists of the determination of a rigid transformation T which brings both images into the same reference system. A crude registration should be independent of data geometry, and be efficient, in the sense that has low time complexity in finding the correspondence pairs. The most popular solution consists of extracting n pairs of correspondences in the overlapping regions and estimating from them the transformation T. Current works on crude registration still present drawbacks in relation to data geometry and efficiency. This work proposes an alternative solution for crude registration. Based on the hypothesis that the shape of an object is essentially captured by a finite subset of its points, this thesis proposes the use of a simplified triangular mesh, instead of a dense range image, to reduce the search space of correspondences. Besides, the construction of a descriptor, called trihedron is proposed. It has higher discriminating capacity which makes correspondence search still more efficient. The proposed algorithm has been validated through experiment tests with several real images (AU) | |