Réalité augmentée automatique à partir d'une séquence vidéo et utilisant la stéréoscopie dense

Papyrus

http://hdl.handle.net/1866/14523

Abstract

This thesis concentrates on augmented reality, which collsists in the integration of virtual objects in real images. This integration must be realized iII a realistic way, which creates a difficulty because of the depth of the real objects. Ail illdustrial ap plications of augmented reality require a strong human illteractivity. The method we propose in this thesis is automatic and only needs a human interaction for the movie capture. First, a dense depth map of the scene is created from the image sequence. After that, we integrate the virtual object, whose depth is known, to the original image where it is in front of the scene elements. This is doue by comparing two depths the depth of the scene and the depth of the virtual object. The dense depth map is created by the max-flow algorithm (ref. [22]), which is combined to two new techniques : the “max-left-right” aud the “depth collsistency”. The “max-left-right” technique solves the haif-occlusions problem. The “depth cousistency” technique of fers a solution to one of the biggest problem of dense stereoscopy the choice of the level of smoothness. In this thesis, ail aspects of our method xviii be discussed, from the movie capture to the virtual object illtegration iII the real images, as well as calibration, stereoscopy, and of course, our two new techniques.

Documents

PDF

---