Recently, creation of realistic 3D contents through sensing the real world has become fundamental for many applications. To enhance 3D geometric models obtained through laser range scanners with their textures reconstructed from several photographic 2D images taken from various view points, it is necessary to determine the camera position and orientation relative to the 3D models for each of the images. In this chapter, a registration method is proposed, which automatically and simultaneously aligns multiple 2D images onto a 3D model. For each iteration process, correspondences between 2D edge pixels and 3D edge points are automatically searched and updated. Besides these 2D-3D edge correspondences, 2D-2D edge correspondences on 3D surface model are also considered simultaneously for global optimization among all the images. Errors are minimized by using conjugate gradient search, utilizing M-estimator for robustness. From texture mapped objects, the usefulness of the proposed simultaneous registration method is shown. Also, it is applied to the creation of digital cultural assets.
All Science Journal Classification (ASJC) codes
- Computer Science(all)