Méthodes de rendu à base de vidéos et applications à la réalité Virtuelle

Abstract : Given a set images of the same scene, the goal of video-based rendering methods is to compute new views of this scene from new viewpoints. The user of this system controls the virtual camera's movement through the scene. Nevertheless, the virtual images are computed from static cameras. A first approach is based on a reconstruction of the scene and can provide accurate models but often requires lengthy computation before visualization. Other methods try to achieve real-time rendering. Our main contribution to video-base rendering concerns the plane sweep method which belongs to the latter family. The plane sweep method divides space in parallel planes. Each point of each plane is processed independently in order to know if it lies on the surface of an object of the scene. These informations are used to compute a new view of the scene from a new viewpoint. This method is well suited to an implementation using graphic hardware and thus to reach realtime rendering. Our main contribution to this method concerns the way to consider whether a point of a plane lies on the surface of an object of the scene. We first propose a new scoring method increasing the visual quality of the new images. Compared with previous approaches, this method implies fewer constraints on the position of the virtaul camera, i.e. this camera does not need to lie between the input camera's area. We also present an adaptation of the plane sweep algorithm that handles partial occlusions. According to video-based rendering practical applications in virtual reality, we propose an improvement of the plane sweep method dealing with stereoscopic images computation that provides visualization of the virtual scene in relief. Our enhancement provides the second view with only low additional computation time whereas most of the others techniques require to render the scene twice. This improvement is based on a sharing of the informations common to the two stereoscopic views. Finally, we propose a method that removes pseudoscopic movements in a virtual reality application. These pseudoscopic movements appear when the observer moves in front of the stereoscopic screen. Then the scene roportions seem to be distorted and the observer sees the objects of the scene moving in an anormal way. The method we propose is available either on a classical stereoscopic rendering method or on the Plane Seep algorithm. Every method we propose widely uses graphic harware through to shader programs and provides real-time rendering. These methods only require a standard computer, a video acquisition device and a powerful enough graphic card. There exists a lot of practicalapplications of the plane sweep method, especially in fields like virtual reality, video games, 3d television or security.
Complete list of metadatas

Contributor : Vincent Nozick <>
Submitted on : Tuesday, September 18, 2012 - 4:54:43 PM
Last modification on : Thursday, July 5, 2018 - 2:25:04 PM
Long-term archiving on : Wednesday, December 19, 2012 - 3:46:08 AM


  • HAL Id : tel-00733470, version 1


Vincent Nozick. Méthodes de rendu à base de vidéos et applications à la réalité Virtuelle. Synthèse d'image et réalité virtuelle [cs.GR]. Université Paris-Est, 2006. Français. ⟨tel-00733470⟩



Record views


Files downloads