Résumé: The emergence of a new generation of 3D auto stereoscopic displays is driving the requirement for multi-baseline images. The dominant form of this display technology requires multiple views of the same scene, captured at a single instance in time along a common baseline in order to project stereoscopic images to the viewer. The direct acquisition of multiple views (typically 8 or 16 for the current generation of such displays) is problematic due to the difficulty of configuring, calibrating and controlling multiple cameras simultaneously. This paper describes a technique that alleviates these problems by generating the required views from binocular images. Considering each stereo pair in isolation leads to inconsistency across image sequences. By incorporating a motion-tracking algorithm this problem is significantly reduced. In this paper we describe a novel approach to stereo matching of image sequences for the purpose of generating multiple virtual camera views. Results of extensive tests on stereo image sequences will be documented indicating that this approach is promising both in terms of the speed of execution and the quality of the results produced.

Langue: Anglais