Multi-user environments, using 3D graphics, more and more find their way into areas like e-business, entertainment and cooperative work. Due to the increasing capabilities of hardware and network, even smaller devices, like laptop computers or PDAs, become suitable for the visualization of 3D graphics. A possible approach for distributing the 3D information is to use a server, which provides all clients with the necessary data. In this paper we define a client-server software framework, which supports the design of three-dimensional multi-user environments. We define requirements of the software infrastructure for a client-server framework that can provide a number of different 3D scenes, each with several users. The number of 3D scenes and users should be only limited by the hardware. Furthermore, we present a solution for each requirement and finally we discuss the results.

This thesis represents a method for reducing the geometry of complex room models to make it possible to model their acoustics in real-time. It is shown that reducing geometry can vastly decrease the modeling time. This conclusion is supported by the experimental results. The previous work on geometric model simplification is surveyed extensively. The re-duction algorithms are grouped into decimation algorithms and surface reconstruction algorithms. The decimation algorithms are categorized into vertex removal, vertex clustering, edge collapsing, and triangle removal algorithms. The surface reconstruction algorithms include re-meshing and volumetric approaches. The main contribution of this thesis is a method for reducing geometric models for acoustics modeling. It consists of two steps: topology simplification and geometry reduction. Two approaches are suggested for the topology simplification, one based on a regular density grid and another based on an octree. There are also two algorithms for the geometry reduction both of which are trying to merge small coplanar triangles into large polygons. The results of the reduction method are evaluated using the acoustical parameters. Also the performance of the reduction algorithm is analyzed. The experiments show that the algorithm preserves spatial properties, measured by the reverberation time, relatively well. The algorithm can produce coarsest level approximations in real-time, but more accurate approximations require significantly longer times for computation.

Titre: Modélisation et distribution adaptatives de grandes scènes naturelles te l-0

Présentée et soutenue par Sebastien Mondet

o.Univ.Prof.Dr.G.Chroust o.Univ.Prof.Dr.G.Haring Univ.Prof.Dr.G.Kotsis Univ.Doz.Dr.V.Risak