Modèles de formation d'image génériques : calibrage et algorithmes de reconstruction 3D
Srikumar Ramalingam · HAL (Le Centre pour la Communication Scientifique Directe) · 2006
Generic Imaging Models: Calibration and 3D Reconstruction algorithms. Vision applications have been using cameras which are beyond pinhole: stereo, fisheye cameras, catadioptric systems, multi-camera setups etc. These novel cameras have interesting properties, especially a large field of view. Camera calibration and 3D reconstruction algorithms are fundamental blocks for computer vision. Models and algorithms for these two problems are usually parametric, camera dependent and seldom capable of handling heterogeneous camera networks, that are useful for complementary advantages. To solve these problems a generic imaging model is introduced, where every camera is modeled as a set of pixels and their associated projection rays. We propose generic methods for calibrating this model, i.e. for computing all these projection rays. These are thus able to calibrate whaetever camera using the same approach. We also propose generic algorithms for structure-from-motion (3D reconstruction, motion and pose estimation, bundle adjustment) and self-calibration.