Video mosaicking for arbitrary scene imaged under arbitrary camera motion

Abstract

Existing mosaic construction algorithms have the following restrictions: (1) the scene is planar or very distant, or (2) the scene is generic but the camera motion is a pure rotation. In either case the registration of images could be captured by a global mapping named planar homography. In this work we address the general case: a generic scene pictured under an arbitrary camera motion. The image data so captured contains parallax that makes the registration of images a difficult problem. We introduce a framework that is capable of constructing mosaic from an image stream of even such a nature. The framework is devised on the basis of an algorithm we refer to as the three-image algorithm that, by the use of a third image, overcomes the parallax problem in registering images. We answer two questions: (1) how an image stream is divided into various 3-image sets for the three-image algorithm to iterate upon; and (2) how intermediate mosaic results over the various 3-image sets are accumulated to compose the overall mosaic at the end. The framework allows uneven sampling, in terms of space or time, of the video stream. Experimental results on real image data are presented to illustrate the performance of the proposed solution. © Springer-Verlag Berlin Heidelberg 2003.