Rapid visualization of large point-based surfaces

  • Authors:
  • Tamy Boubekeur;Florent Duguet;Christophe Schlick

  • Affiliations:
  • LaBRI - INRIA - CNRS - University of Bordeaux;INRIA Sophia Antipolis - ENST Paris;LaBRI - INRIA - CNRS - University of Bordeaux

  • Venue:
  • VAST'05 Proceedings of the 6th International conference on Virtual Reality, Archaeology and Intelligent Cultural Heritage
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

Point-Based Surfaces can be directly generated by 3D scanners and avoid the generation and storage of an explicit topology for a sampled geometry, which saves time and storage space for very dense and large objects, such as scanned statues and other archaeological artefacts [DDGM]. We propose a fast processing pipeline of large point-based surfaces for real-time, appearance preserving, polygonal rendering. Our goal is to reduce the time needed between a point set made of hundred of millions samples and a high resolution visualization taking benefit of modern graphics hardware, tuned for normal mapping of polygons. Our approach starts by an out-of-core generation of a coarse local triangulation of the original model. The resulting coarse mesh is enriched by applying a set of maps which capture the high frequency features of the original data set. We choose as an example the normal component of samples for these maps, since normal maps provide efficiently an accurate local illumination. But our approach is also suitable for other point attributes such as color or position (displacement map). These maps come also from an out-of-core process, using the complete input data in a streaming process. Sampling issues of the maps are addressed using an efficient diffusion algorithm in 2D. Our main contribution is to directly handle such large unorganized point clouds through this two pass algorithm, without the time-consuming meshing or parameterization step, required by current state-of-the-art high resolution visualization methods. One of the main advantages is to express most of the fine features present in the original large point clouds as textures in the huge texture memory usually provided by graphics devices, using only a lazy local parameterization. Our technique comes as a complementary tool to high-quality, but costly, out-of-core visualization systems. Direct applications are: interactive preview at high screen resolution of very detailed scanned objects such as scanned statues, inclusion of large point clouds in usual polygonal 3D engines and 3D databases browsing.