Eye Gaze-Guided Matching and Reconstruction of 3-D Model
| Autor: | Chih-Chin Yang, 楊智欽 |
|---|---|
| Rok vydání: | 2003 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 91 Photo-realistic reconstruction of a 3-D architecture scene has been a very important research topic in the field of computer vision. 3-D reconstruction techniques are emerging to many applications of millitary, recreation, and commercial applications. Among all the 3-D reconstruction techniques, stereo vision has been considered to be the most important one. However, because the stereo correspondence problem is still an open question, it is inevitable to use interactive methods to solve for the stereo correspondence problem. In this work, a novel stereo interface is developed specifically for interactive 3-D reconstruction. The stereo interface consists of a gaze tracking module, an eye-gaze guided 3-D reconstruction module, and a graphical user interface. The eye-gaze guided 3-D reconstruction module will be described in detail in this thesis. In order to reconstruct a 3-D architecture with uncalibrated images, a camera self-calibration method is implemented to compute the camera parameters of the uncalibrated stereo images. The architecture reconstruction problem is divided into many sub-problems of reconstructing building blocks such as cuboids, roofs, cylinders, and spheres. The operator selects proper building blocks according to the stereo images of a architecture to reconstruct its 3-D model. To obtain the parameters of a building block, the gaze tracker records the trajectories of the binocular fixation points of the operator while the operator stares at several feature points on the surface of the building block. Then, the system automatically compute the stereo correspondences based on the binocular trajectories of the fixation points of the operator. Once the stereo correspondences are computed, 3-D coordinates of the feature points can be evaluated using the stereo triangulation technique. The parameters of each surface of the building block can then be estimated using the RANSAC algorithm. Experimental results show that the eye-gaze guided 3-D reconstruction module is very promising. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|