Calibration is the first step in image augmentation. Classical approaches compute the projection matrix given 3D points of the scene and their 2D image correspondences. Different auto-calibration algorithms have been recently developed by the computer vision community. They do not use 3D-2D correspondences, but need many 2D-2D correspondences over long sequence of images to provide stable results. In this article we propose a calibration propagation procedure, which is in-between the two previous approaches.Starting from one calibrated image, the unknown camera parameters and position are computed for a second image. In particular, this paper presents a method for extracting the focal length and the 3D structure, while other camera intrinsic parameters remain invariant. In practice for many professional cameras the principal point is approximately at the center of the image and the aspect ratio is given by camera specification.Calibration propagation is relevant to augmented reality applications, e.g. video see through HMD with zooming capability, since it enables image augmentation for a number of camera views with changing intrinsic parameters. In this paper, we present results on synthetic images showing the theoretical validity and performance of the method. We then use real data to demonstrate the potential of this approach for image augmentation applications in industrial maintenance assistance and architectural design.     «

Calibration is the first step in image augmentation. Classical approaches compute the projection matrix given 3D points of the scene and their 2D image correspondences. Different auto-calibration algorithms have been recently developed by the computer vision community. They do not use 3D-2D correspondences, but need many 2D-2D correspondences over long sequence of images to provide stable results. In this article we propose a calibration propagation procedure, which is in-between the two previou...     »