BROWSE
Detail
Conference Paper
Implementation of 3D Structure Reconstruction System from Multiple Images
Cited - time in 
Share 
Share
- Issue Date
- 2006-02
- Citation
- Korea-Japan Joint Workshop on Frontiers of Computer Vision (FCV) 2006, pp.273-277
- Language
- English
- Type
- Conference Paper
- Abstract
- This paper presents an image-based 3D reconstruction system implemented at ETRI. Our system consists of three modules: camera self-calibration, object modeling and texture extraction. For camera self-calibration, we provide three kinds of self-calibration methods from which users can select the appropriate one according to their requirement. Modeling module first calculates 3D coordinates of the objects from user-defined point correspondences. And then, both polyhedral objects and curved objects are reconstructed from the calibrated 3D coordinates. In order to model curved objects, it employs NURBS curves and surfaces. As a final step, it extracts textures from images. To remove perspective distortions and illumination changes between input images, we implement a view-dependent texture extraction algorithm based on homography matrix. Using this system, it is possible to reconstruct a various type of polyhedral and curved objects without much labor. Experimental results show the effectiveness of our system.
- KSP Keywords
- 3D Coordinates, 3D structure, Calibration method, Camera Self-calibration, Curves and surfaces, Illumination change, Image-based 3D reconstruction, NURBS curves, Point correspondences, Texture extraction, User-defined