Using mosaic images to simplify the integration of virtual objects into image sequences

Abstract

This article presents an automatic system to integrate virtual objects into image sequences taken by a rotating and zooming camera. In this case of camera motion the observed static scene can be represented as a 2D panoramic mosaic image, which is estimated from the image sequence. Virtual objects can be positioned easily in the mosaic image with standard image-editing software. The modifications made on the mosaic image are applied to all images of the sequence by the system. Occlusions of virtual objects by moving foreground objects of the real sequence are taken into account. In contrast to existing computer vision approaches, the proposed system is very easy to handle, because no 3D modelling and animation tool is required for the positioning of virtual objects in the 3D geometry of the observed scene. If moving foreground objects from the real sequence occlude virtual objects, usually the required image masks have to be generated manually. The presented system calculates these masks automatically.