Integration of Virtual Objects into an Image from a Single View : Geometrical Properties of Images and Panoramas

The aim of this work is the study of the integration of virtual objects into single views. The use of single views in augmented reality is much more flexible for the end-user than stereo pairs of images, because they require less intervention and preparation. A fundamental problem of augmented reality is to accurately recover the needed camera parameters. While this problem is well-known when two views are available, specific algorithms must be considered for the case of single views. These algorithms generally make use of the geometric particularities of the scene. The camera model and its geometry are therefore of great importance, and are largely analyzed in this report. The first task is to calibrate the camera by finding its internal parameters (mainly the focal length and the principal point). One method for it is the use of vanishing points. Techniques for the recovery of vanishing points in an image are presented, as well as known calibration methods based on it. Another way to calibrate the camera is to use planar structures, and an algorithm based on multiple planes is derived. The special case of panorama is also discussed, with details about its construction and ist calibration. The second task for the integration of virtual objects is the estimation of the camera position and orientation relative to a given coordinate system (pose). Again, vanishing points and plane-based methods are discussed. Two novel algorithms for pose estimation from multiple planes are presented. The pose can also be estimated in the case of known 3D points coordinates. Seven recent algorithms based on this assumption are detailed and compared. All the algorithms mentioned in this work have been implemented as a practical goal, and simple objects could be integrated into various images using different techniques. The report discusses the implementation particularities and presents the results obtained by comparative tests