Point Location

In this tutorial, we demonstrate how triangulations can be used to perform point location. The problem of interest is: Given a point p and a triangulation tri, what triangle T in tri contains p? We provide a function find_triangle for this task, implementing the algorithm of Mücke, Saias, and Zhu (1999). The algorithm has been slightly modified to allow for regions with holes. Support is also provided for non-convex and disjoint domains, but the algorithm is significantly slower in these cases and requires some special case. (The approach for these cases is, basically, to just keep trying new points to start the algorithm from until it works, but you the user must specify a keyword argument concavity_protection to make an extra check to guarantee even greater safety.)

Unconstrained example

We start with a simple example, demonstrating point location on an unconstrained triangulation.

using DelaunayTriangulation
using CairoMakie
using StableRNGs

points = [
    (-3.0, 6.0), (5.0, 1.0), (-5.0, 3.0), (2.0, -3.0),
    (5.0, 8.0), (0.0, 0.0), (2.0, 5.0), (-3.0, 1.0),
    (-2.0, -1.0), (-1.0, 4.0),
]
tri = triangulate(points)
q = (3.0, 3.0)
fig, ax, sc = triplot(tri)
scatter!(ax, q)
fig

The aim is to, from tri, find which triangle contains the point q shown. Using the find_triangle function, this is simple.

V = find_triangle(tri, q)

(2, 7, 6)

The result means that the triangle (2, 7, 6) contains the point, as we can easily check:

DelaunayTriangulation.point_position_relative_to_triangle(tri, V, q)

Certificate.Inside = 0

When we provide no keyword arguments, the default behaviour of find_triangle is to first sample some number of points (defaults to $\lceil \sqrt[3]{n}\rceil$, where $n$ is the number of points), and then start at the point that is closest to q out of those sampled, then marching along the triangulation until q is found. This number of samples can be changed using the m keyword argument. For example,

V = find_triangle(tri, q, m = 10)

(6, 2, 7)

means that we get a sample of size 10, and start at whichever point is the closest. (For technical reasons, this sampling is with replacement, so it is possible that the same point is sampled more than once.) You could also instead specify the point to start at using the k keyword argument, in which case no points are sampled. For example,

V = find_triangle(tri, q, k = 6)

(2, 7, 6)

starts the algorithm at the point 6.

Note also that the triangles found from find_triangle do not have to be given in the same order as they appear in the triangulation. For example, if a triangle (i, j, k) contains the point q, then any of (i, j, k), (j, k, i), or (k, i, j) could be returned.

The point q does not have to be in the triangulation. For example, consider the following point.

q = (-5.0, 8.0)
fig, ax, sc = triplot(tri)
scatter!(ax, q)
fig

We obtain:

V = find_triangle(tri, q)

(1, 5, -1)

See that the result is a ghost triangle (1, 5, -1). As discussed in the manual, this can be interpreted as meaning that q is between the two lines through the points 1 and 5 that start at a central point of the triangulation. (The index -1 is just the ghost vertex.) This can be visualised.

fig, ax, sc = triplot(tri, show_ghost_edges = true)
scatter!(ax, q)
fig

Region with concave boundaries and holes

Now we give an example of point location for a reason with holes. Since the case where all boundaries are convex is reasonably straight forward, here we consider concave boundaries and discuss methods for improving the speed of the algorithm in this case. First, let us give our example triangulation.

a, b, c = (0.0, 8.0), (0.0, 6.0), (0.0, 4.0)
d, e, f = (0.0, 2.0), (0.0, 0.0), (2.0, 0.0)
g, h, i = (4.0, 0.0), (6.0, 0.0), (8.0, 0.0)
j, k, ℓ = (8.0, 1.0), (7.0, 2.0), (5.0, 2.0)
m, n, o = (3.0, 2.0), (2.0, 3.0), (2.0, 5.0)
p, q, r = (2.0, 7.0), (1.0, 8.0), (1.0, 2.2)
s, t, u = (0.4, 1.4), (1.2, 1.8), (2.8, 0.6)
v, w, z = (3.4, 1.2), (1.6, 1.4), (1.6, 2.2)
outer = [[a, b, c, d, e], [e, f, g, h, i, j, k, ℓ], [ℓ, m, n, o, p, q, a]]
inner = [[r, z, v, u, w, t, s, r]]
boundary_nodes, points = convert_boundary_points_to_indices([outer, inner])
rng = StableRNG(125123)
tri = triangulate(points; rng, boundary_nodes)
refine!(tri; max_area = 0.01get_area(tri), rng);

The issue with concavity is that the ghost triangles can no longer be sensibly defined. To demonstrate this, see the following plot:

fig, ax, sc = triplot(tri, show_ghost_edges = true)
fig

The ghost edges now intersect the boundary, which doesn't make sense, and creates difficulties. Let us now demonstrate how the function still works here. We try finding the blue points shown below.

qs = [
    (4.0, 5.0), (1.0, 5.6), (0.2, 5.0),
    (0.0, -1.0), (0.5, 3.5), (2.5, 1.5),
    (1.0, 2.0), (4.5, 1.0), (6.0, 1.5),
    (0.5, 8.5), (1.0, 7.5), (1.2, 1.6),
]
fig, ax, sc = triplot(tri, show_ghost_edges = false)
scatter!(ax, qs, color = :blue, markersize = 16)
fig

Now let's find the triangles.

Vs = [find_triangle(tri, q; rng) for q in qs]

12-element Vector{Tuple{Int64, Int64, Int64}}:
 (35, 12, -3)
 (76, 56, 75)
 (83, 55, 73)
 (68, 6, -2)
 (103, 47, 88)
 (54, 25, -4)
 (18, 24, -4)
 (70, 59, 36)
 (62, 53, 61)
 (96, 17, -3)
 (79, 17, 80)
 (-4, 18, 24)

While we do find some triangles, they may not all be correct. For example, the triangle found for (1.2, 1.6) is

Vs[end]

(-4, 18, 24)

but the point (1.2, 1.6) is actually inside the triangulation. To protect against this, you need to use concavity_protection=true, which will enable a check to be made that the point is actually outside the triangulation whenever a ghost triangle is to be returned. If the check finds this to not be the case, it restarts. With these results, we now compute:

Vs = [find_triangle(tri, q; rng, concavity_protection = true) for q in qs]

12-element Vector{Tuple{Int64, Int64, Int64}}:
 (35, 12, -3)
 (75, 76, 56)
 (55, 73, 83)
 (68, 6, -2)
 (88, 103, 47)
 (25, -4, 54)
 (18, 24, -4)
 (57, 65, 36)
 (62, 53, 61)
 (96, 17, -3)
 (17, 80, 79)
 (22, 23, 41)

Here is how we can actually test that these results are now correct. We cannot directly use DelaunayTriangulation.point_position_relative_to_triangle because it does not know that the ghost triangles are invalid. Instead, we find the distance of each point to the triangulation's boundary using DelaunayTriangulation.dist so that we can classify it as being inside or outside of the triangulation, and then check the type of the found triangle.

δs = [DelaunayTriangulation.dist(tri, q) for q in qs]
results = Vector{Bool}(undef, length(qs))
for (j, (q, δ, V)) in (enumerate ∘ zip)(qs, δs, Vs)
    cert = DelaunayTriangulation.point_position_relative_to_triangle(tri, V, q)
    is_ghost = DelaunayTriangulation.is_ghost_triangle(V)
    is_outside = DelaunayTriangulation.is_outside(cert)
    if δ ≥ 0.0
        results[j] = !is_outside && !is_ghost
    else # δ < 0.0 ⟹ outside
        results[j] = !is_outside && is_ghost
    end
end
results

12-element Vector{Bool}:
 1
 1
 1
 1
 1
 1
 1
 1
 1
 1
 1
 1

As we see, the triangles are now all correct.

Disjoint domains

Now we continue the previous example by adding in another set of domains that are disjoint to the current domain, thus allowing us to demonstrate how find_triangle applies here. The new domain is below, along with the points we will be searching for.

m₁, n₁, o₁ = (6.0, 8.0), (8.0, 8.0), (8.0, 4.0)
p₁, q₁, r₁ = (10.0, 4.0), (6.0, 6.0), (8.0, 6.0)
s₁, t₁, u₁ = (9.0, 7.0), (4.0, 4.0), (5.0, 4.0)
v₁, w₁ = (5.0, 3.0), (4.0, 3.0)
new_domain₁ = [[m₁, q₁, o₁, p₁, r₁, s₁, n₁, m₁]]
new_domain₂ = [[t₁, w₁, v₁, u₁, t₁]]
boundary_nodes, points = convert_boundary_points_to_indices(
    [outer, inner, new_domain₁, new_domain₂],
)
rng = StableRNG(125123)
tri = triangulate(points; rng, boundary_nodes)
refine!(tri; max_area = 0.001get_area(tri), rng)
qs = [
    (0.6, 6.4), (1.4, 0.8), (3.1, 2.9),
    (6.3, 4.9), (4.6, 3.5), (7.0, 7.0),
    (8.9, 5.1), (5.8, 0.8), (1.0, 1.5),
    (1.5, 2.0), (8.15, 6.0),
]
fig, ax, sc = triplot(tri)
scatter!(ax, qs, color = :blue, markersize = 16)
fig

Here are the find_triangle results.

Vs = [find_triangle(tri, q; rng, concavity_protection = true) for q in qs]

11-element Vector{Tuple{Int64, Int64, Int64}}:
 (412, 357, 378)
 (425, 811, 424)
 (47, 172, -3)
 (598, 828, -5)
 (586, 595, 944)
 (665, 774, 604)
 (579, 721, 735)
 (117, 816, 920)
 (818, 857, 237)
 (18, -4, 294)
 (88, 142, -2)

Again, we can verify that these are all correct as follows. Without concavity_protection=true, these would not be all correct.

δs = [DelaunayTriangulation.dist(tri, q) for q in qs]
results = Vector{Bool}(undef, length(qs))
for (j, (q, δ, V)) in (enumerate ∘ zip)(qs, δs, Vs)
    cert = DelaunayTriangulation.point_position_relative_to_triangle(tri, V, q)
    is_ghost = DelaunayTriangulation.is_ghost_triangle(V)
    is_outside = DelaunayTriangulation.is_outside(cert)
    if δ ≥ 0.0
        results[j] = !is_outside && !is_ghost
    else # δ < 0.0 ⟹ outside
        results[j] = !is_outside && is_ghost
    end
end
results

11-element Vector{Bool}:
 1
 1
 1
 1
 1
 1
 1
 1
 1
 1
 1

Just the code

An uncommented version of this example is given below. You can view the source code for this file here.

using DelaunayTriangulation
using CairoMakie
using StableRNGs

points = [
    (-3.0, 6.0), (5.0, 1.0), (-5.0, 3.0), (2.0, -3.0),
    (5.0, 8.0), (0.0, 0.0), (2.0, 5.0), (-3.0, 1.0),
    (-2.0, -1.0), (-1.0, 4.0),
]
tri = triangulate(points)
q = (3.0, 3.0)
fig, ax, sc = triplot(tri)
scatter!(ax, q)
fig

V = find_triangle(tri, q)

DelaunayTriangulation.point_position_relative_to_triangle(tri, V, q)

V = find_triangle(tri, q, m = 10)

V = find_triangle(tri, q, k = 6)

q = (-5.0, 8.0)
fig, ax, sc = triplot(tri)
scatter!(ax, q)
fig

V = find_triangle(tri, q)

fig, ax, sc = triplot(tri, show_ghost_edges = true)
scatter!(ax, q)
fig

a, b, c = (0.0, 8.0), (0.0, 6.0), (0.0, 4.0)
d, e, f = (0.0, 2.0), (0.0, 0.0), (2.0, 0.0)
g, h, i = (4.0, 0.0), (6.0, 0.0), (8.0, 0.0)
j, k, ℓ = (8.0, 1.0), (7.0, 2.0), (5.0, 2.0)
m, n, o = (3.0, 2.0), (2.0, 3.0), (2.0, 5.0)
p, q, r = (2.0, 7.0), (1.0, 8.0), (1.0, 2.2)
s, t, u = (0.4, 1.4), (1.2, 1.8), (2.8, 0.6)
v, w, z = (3.4, 1.2), (1.6, 1.4), (1.6, 2.2)
outer = [[a, b, c, d, e], [e, f, g, h, i, j, k, ℓ], [ℓ, m, n, o, p, q, a]]
inner = [[r, z, v, u, w, t, s, r]]
boundary_nodes, points = convert_boundary_points_to_indices([outer, inner])
rng = StableRNG(125123)
tri = triangulate(points; rng, boundary_nodes)
refine!(tri; max_area = 0.01get_area(tri), rng);

fig, ax, sc = triplot(tri, show_ghost_edges = true)
fig

qs = [
    (4.0, 5.0), (1.0, 5.6), (0.2, 5.0),
    (0.0, -1.0), (0.5, 3.5), (2.5, 1.5),
    (1.0, 2.0), (4.5, 1.0), (6.0, 1.5),
    (0.5, 8.5), (1.0, 7.5), (1.2, 1.6),
]
fig, ax, sc = triplot(tri, show_ghost_edges = false)
scatter!(ax, qs, color = :blue, markersize = 16)
fig

Vs = [find_triangle(tri, q; rng) for q in qs]

Vs[end]

Vs = [find_triangle(tri, q; rng, concavity_protection = true) for q in qs]

δs = [DelaunayTriangulation.dist(tri, q) for q in qs]
results = Vector{Bool}(undef, length(qs))
for (j, (q, δ, V)) in (enumerate ∘ zip)(qs, δs, Vs)
    cert = DelaunayTriangulation.point_position_relative_to_triangle(tri, V, q)
    is_ghost = DelaunayTriangulation.is_ghost_triangle(V)
    is_outside = DelaunayTriangulation.is_outside(cert)
    if δ ≥ 0.0
        results[j] = !is_outside && !is_ghost
    else # δ < 0.0 ⟹ outside
        results[j] = !is_outside && is_ghost
    end
end
results

m₁, n₁, o₁ = (6.0, 8.0), (8.0, 8.0), (8.0, 4.0)
p₁, q₁, r₁ = (10.0, 4.0), (6.0, 6.0), (8.0, 6.0)
s₁, t₁, u₁ = (9.0, 7.0), (4.0, 4.0), (5.0, 4.0)
v₁, w₁ = (5.0, 3.0), (4.0, 3.0)
new_domain₁ = [[m₁, q₁, o₁, p₁, r₁, s₁, n₁, m₁]]
new_domain₂ = [[t₁, w₁, v₁, u₁, t₁]]
boundary_nodes, points = convert_boundary_points_to_indices(
    [outer, inner, new_domain₁, new_domain₂],
)
rng = StableRNG(125123)
tri = triangulate(points; rng, boundary_nodes)
refine!(tri; max_area = 0.001get_area(tri), rng)
qs = [
    (0.6, 6.4), (1.4, 0.8), (3.1, 2.9),
    (6.3, 4.9), (4.6, 3.5), (7.0, 7.0),
    (8.9, 5.1), (5.8, 0.8), (1.0, 1.5),
    (1.5, 2.0), (8.15, 6.0),
]
fig, ax, sc = triplot(tri)
scatter!(ax, qs, color = :blue, markersize = 16)
fig

Vs = [find_triangle(tri, q; rng, concavity_protection = true) for q in qs]

δs = [DelaunayTriangulation.dist(tri, q) for q in qs]
results = Vector{Bool}(undef, length(qs))
for (j, (q, δ, V)) in (enumerate ∘ zip)(qs, δs, Vs)
    cert = DelaunayTriangulation.point_position_relative_to_triangle(tri, V, q)
    is_ghost = DelaunayTriangulation.is_ghost_triangle(V)
    is_outside = DelaunayTriangulation.is_outside(cert)
    if δ ≥ 0.0
        results[j] = !is_outside && !is_ghost
    else # δ < 0.0 ⟹ outside
        results[j] = !is_outside && is_ghost
    end
end
results

This page was generated using Literate.jl.