@desicochrane/delaunay v0.1.2

Delaunay

Dependency-free implementation of the divide-and-conquer Delaunay Triangulation algorithm.

Getting started

1. npm i @desicochrane/delaunay

2. import and triangulate your point set:

   import Delaunay, {UniqueEdges} from '@desicochrane/delaunay'
   
   // define your point set
   const pts = [ [1,1], [1,2], [3,4], [4,5] ]
   
   // compute triangulation graph
   const graph = Delaunay(pts)
   
   // get unique edges from graph
   const edges = UniqueEdges(graph)

The Algorithm

This package takes a divide-and-conquer approach in the same flavor as merge-sort, the basic idea goes something like:

1. Sort the points left to right
2. Divide the points into halves until point set has size 2 or 3
3. Triangulate the 2 or 3 point sets manually
4. Merge the triangulations together

A sample code of the main routine would look something like this:

function Delaunay(pts) {
    // the output will be an adjacency list of edges
    const edges = {}
    
    // pass edges into the recursive procedure
    delaunay(edges, pts.sort(PointSort), 0, pts.length-1)
    
    return edges
}

function delaunay(edges, pts, min, max) {
    // how many points are we triangulating
    const size = max-min+1
    
    // zero or 1 point is the trivial delaunay triangulation 
    if (size < 1) return edges
    
    // 2 points can be computed easily
    if (size == 2) return Triangulate2(edges, pts[min], pts[max])
    
    // 3 points can be computed easily
    if (size == 3) return Triangulate3(edges, pts[min], pts[min+1], pts[max])
    
    // otherwise we divide into halves:
    const mid = min+((max-min)>>>1)
    
    delaunay(edges, pts, min, mid)
    delaunay(edges, pts, mid+1, max)
    
    // and then merge the results
    return Merge(adj, mid, mid+1)
}

What remains then is to specify the subroutines PointSort, Triangulate2, Triangulate3, and Merge.

PointSort Subroutines

todo

Triangulate2+3 Subroutines

todo

Merge Subroutine

todo

Geometry Prerequisites

The implementation depends on the following geometric concepts and properties:

PseudoAngle

todo

RightOf

todo

Circumscribed

todo

Convex Hull

todo

Data Structures

The efficiency depends heavily on being able to traverse the triangulation efficiently, to do so we construct a Vertex Cycle data structure which is effectively a specialised Cyclic Linked List.

Cyclic Linked List

todo

Vertex Cycle

todo