PrusaSlicer-NonPlainar/src/qhull/Announce.txt
bubnikv 0558b53493 WIP: Moved sources int src/, separated most of the source code from Perl.
The XS was left only for the unit / integration tests, and it links
libslic3r only. No wxWidgets are allowed to be used from Perl starting
from now.
2018-09-19 11:02:24 +02:00

48 lines
2.1 KiB
Plaintext

Qhull 2015.2 2016/01/18
http://www.qhull.org
git@github.com:qhull/qhull.git
http://www.geomview.org
Qhull computes convex hulls, Delaunay triangulations, Voronoi diagrams,
furthest-site Voronoi diagrams, and halfspace intersections about a point.
It runs in 2-d, 3-d, 4-d, or higher. It implements the Quickhull algorithm
for computing convex hulls. Qhull handles round-off errors from floating
point arithmetic. It can approximate a convex hull.
The program includes options for hull volume, facet area, partial hulls,
input transformations, randomization, tracing, multiple output formats, and
execution statistics. The program can be called from within your application.
You can view the results in 2-d, 3-d and 4-d with Geomview.
To download Qhull:
http://www.qhull.org/download
git@github.com:qhull/qhull.git
Download qhull-96.ps for:
Barber, C. B., D.P. Dobkin, and H.T. Huhdanpaa, "The
Quickhull Algorithm for Convex Hulls," ACM Trans. on
Mathematical Software, 22(4):469-483, Dec. 1996.
http://www.acm.org/pubs/citations/journals/toms/1996-22-4/p469-barber/
http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.117.405
Abstract:
The convex hull of a set of points is the smallest convex set that contains
the points. This article presents a practical convex hull algorithm that
combines the two-dimensional Quickhull Algorithm with the general dimension
Beneath-Beyond Algorithm. It is similar to the randomized, incremental
algorithms for convex hull and Delaunay triangulation. We provide empirical
evidence that the algorithm runs faster when the input contains non-extreme
points, and that it uses less memory.
Computational geometry algorithms have traditionally assumed that input sets
are well behaved. When an algorithm is implemented with floating point
arithmetic, this assumption can lead to serious errors. We briefly describe
a solution to this problem when computing the convex hull in two, three, or
four dimensions. The output is a set of "thick" facets that contain all
possible exact convex hulls of the input. A variation is effective in five
or more dimensions.