Hardware-Accelerated Rendering of Unprocessed Point Clouds

Information

  • Publication Type: Master Thesis
  • Workgroup(s)/Project(s):
  • Date: 2006
  • First Supervisor: Michael Wimmer
  • Keywords: out-of-core rendering, point-based rendering, datastructure

Abstract

In this diploma thesis a fast rendering algorithm for very large point clouds is described. A point cloud is simply a set of unconnected 3D coordinates in cartesian space. Each coordinate of such a set is interpreted as a point in space. A point cloud is the result of a sampling process, where either a laser scanner samples a real environment, or the data structure of some already existing graphical model is point sampled. During rendering it is attempted to reconstruct the sampled model from the given point cloud. The algorithm presented in this thesis builds on two new data structures, namely Memory Optimized Sequential Point Trees and Nested Octrees. It includes an out-of-core part, which means that it is also possible to render models that do not fit in the main memory of the computer, and an occlusion-culling part, which means that objects, which are hidden by objects closer to the viewer, do not have to be rendered. The algorithm is developed primarily for the fast rendering of point clouds, i.e., with a high frame rate, whereas the visual quality of the rendered point clouds is not in the focus of this work. The algorithm does not need any additional attributes at a point besides the position.

Additional Files and Images

Additional images and videos

image: Stephansdom from outside image:Stephansdom from outside
image2: Stephansdom from inside image2:Stephansdom from inside

Additional files

thesis: PDF of the diploma thesis thesis:PDF of the diploma thesis

Weblinks

No further information available.

BibTeX

@mastersthesis{Scheiblauer-2006-DA,
  title =      "Hardware-Accelerated Rendering of Unprocessed Point Clouds",
  author =     "Claus Scheiblauer",
  year =       "2006",
  abstract =   "In this diploma thesis a fast rendering algorithm for very
               large point clouds is described. A point cloud is simply a
               set of unconnected 3D coordinates in cartesian space. Each
               coordinate of such a set is interpreted as a point in space.
               A point cloud is the result of a sampling process, where
               either a laser scanner samples a real environment, or the
               data structure of some already existing graphical model is
               point sampled. During rendering it is attempted to
               reconstruct the sampled model from the given point cloud.
               The algorithm presented in this thesis builds on two new
               data structures, namely Memory Optimized Sequential Point
               Trees and Nested Octrees. It includes an out-of-core part,
               which means that it is also possible to render models that
               do not fit in the main memory of the computer, and an
               occlusion-culling part, which means that objects, which are
               hidden by objects closer to the viewer, do not have to be
               rendered. The algorithm is developed primarily for the fast
               rendering of point clouds, i.e., with a high frame rate,
               whereas the visual quality of the rendered point clouds is
               not in the focus of this work. The algorithm does not need
               any additional attributes at a point besides the position.",
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology",
  keywords =   "out-of-core rendering, point-based rendering, datastructure",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2006/Scheiblauer-2006-DA/",
}