This paper describes an environment to automatically or semi-automatically compute the precise mapping between a set of 2D images and a triangulated 3D model built from high-resolution 3D range data. This environment is part of our Atelier3D framework for the modeling, visualization and analysis of large sensor-based datasets. This work was done to initially support three cultural heritage application projects: the modeling of the Grotta dei Cervi in Italy, of the Erechtheion in Athens, Greece, and of Leonardo's Mona Lisa. The proposed method combines image-based registration, feature matching, robust estimation techniques and advanced multi-resolution rendering with a powerful user interface.

This paper presents a new technique for fast, view-dependent, real-time visualization of large multiresolution geometric models with color or texture information. This method uses geomorphing to smoothly interpolate between geometric patches composing a hierarchical level-of-detail structure, and to maintain seamless continuity between neighboring patches of the model. It combines the advantages of view-dependent rendering with numerous additional features: the high performance rendering associated with static preoptimized geometry, the capability to display at both low and high resolution with minimal artefacts, and a low CPU usage since all the geomorphing is done on the GPU. Furthermore, the hierarchical subdivision of the model into a tree structure can be accomplished according to any spatial or topological criteria. This property is particularly useful in dealing with models with high resolution textures derived from digital photographs. Results are presented for both highly tesselated models (372 million triangles), and for models which also contain large quantities of texture (200 million triangles + 20 GB of compressed texture). The method also incorporates asynchronous out-of-core model management. Performances obtained on commodity hardware are in the range of 50 million geomorphed triangles/second for a benchmark model such as Stanford's St. Matthew dataset.

We present a new method for the stereoscopic display of complex virtual environments using a foveated arrangement of four images. The system runs on four rendering nodes and four projectors, for the fovea and periphery in each eye view. The use of high-resolution insets in a foveated configuration is well known. However, its extension to projector-based stereoscopic displays raises a specific issue: the visible boundary between fovea and periphery present in each eye creates a stereoscopic cue that may conflict with the perceived depth of the underlying scene. A previous solution to this problem displaces the boundary in the images to ensure that it is always positioned over stereoscopically corresponding scene locations. The new method proposed here addresses the same problem, but by relaxing the stereo matching criteria and reformulating the problem as one of spatial partitioning, all computations are performed locally on each node, and require a small and fixed amount of post-rendering processing, independent of scene complexity. We discuss this solution and present an OpenGL implementation; we also discuss acceleration techniques using culling and fragments, and illustrate the use of the method on a complex 3D textured model of a Byzantine crypt built using laser range imaging and digital photography.

No abstract available.

VR systems were traditionally used for tasks relying on high-quality graphics rendering where virtual environments were entirely made of user-defined objects. We are foreseeing that serious breakthroughs will emerge, where 'augmented reality' environments will be created in a more dynamic fashion, using 2D and 3D data from computer vision sensors. Thus, virtual environments will not be made exclusively of user-defined objects, but also from real data. This data, after proper modeling to provide some behaviour and abstraction levels, will then be used to feed high-performance imaging systems. This paper deals with: 1) local surface modeling of the 3D data visible from each 2D viewpoint using a modified marching cubes algorithm; and 2) region matching of the local models (from neighboring views in the sequence) using color region clustering information from the 2D snapshots. Local models and the region matching information are used to build complete global models from a sequence of images.