Talk:Ray-casting algorithm

Fortran: The point of pts?

What is pts used for in the Fortran example? --Paddy3118 01:17, 26 May 2009 (UTC)

Ah, It seems there are one set of points for all polygons. --Paddy3118 01:41, 26 May 2009 (UTC)
Yes, I preferred to use a single set, and to index it (it's not an odd technics if a subset of points is shared among several polygons we want to define). --ShinTakezou 16:53, 26 May 2009 (UTC)

AutoHotkey version

The AutoHotkey solution refers to ray_intersects_segment without defining it, but it seems to me that defining ray_intersects_segment is a key part of this task. --Mr2001 07:31, 31 October 2010 (UTC)

Help:ENA Templates will be helpful for flagging this kind of thing. --Michael Mol 13:23, 31 October 2010 (UTC)

"Coherent results"

The C example claims to "reveal the meaning of coherent results", whereby it considers a point lying on the left edge of a square outside, but a point on right side inside. This must be a meaning of "coherent" that I wasn't aware of. Also, the intersection code is terrible: it uses an arbitrarily defined value as a floating point precision limit while not rigorously enforcing it, which will cause all kinds of unexpected behaviors when used in real work. --Ledrug 22:10, 25 July 2011 (UTC)

You clearly failed to understand the actual meaning and also to cite correctly the text (this is italics on purpose). The text under C code correctly claimed to "reveal the meaning of coherent results". It referred to the text of the problem, where it was and it is stated: Points on the boundary or "on" a vertex are someway special and through this approach we do not obtain coherent results. What does it mean that we don't obtain coherent results? The C code showed it, as expected and anticipated by through this approach, which was clearly the approach used in the implementation. So, it wasn't a dictionary problem on my part, but an understanding problem on your part.
There was a second incredible misunderstanding: in the "original" C code, there wasn't anything to deal with a floating point precision limit, hence it was impossible such a limit was enforced rigorously (or not) anywhere. Clearly you failed to understand what the code did — and this is fine, maybe it was sloppy and messy to you. Nonetheless, the text of the problem says To avoid the "ray on vertex" problem, the point is moved upward of a small quantity ε. Maybe you missed this part, or you believed that every ε/eps/epsilon must be used to enforce a floating point precision limit.
Instead in these cases (check other examples) ε/eps/epsilon is just the value we shift a point in order to simplistically cope with a specific problem. Not the best we can do, and it has "side effects": already considered in the text, thus no surprise (except for you, it seems).
If we need to be pedantic and focus on all the details for a production ready implementation (which isn't the main aim of this wiki, as far as I understand it), floating point should be treated carefully in every comparisons, but many did the naive approach — I think because it isn't the point of the task (or maybe because we hope the target CPU has an instruction like FCMPE in MMIX — just joking). --- ShinTakezou (talk) 18:59, 23 February 2018 (UTC)

D code generate false positive

We port the D code in our C++ project and we found false positive, I almost sure all other implementations could have exactly the same issue.

Here is my modifications of the D code showing the issue (show comments) : <lang d> import std.stdio, std.math, std.algorithm;

immutable struct Point { float x, y; } // Using float instead of double (reducing precision type) immutable struct Edge { Point a, b; } immutable struct Figure {

   string name;
   Edge[] edges;

}

bool contains(in Figure poly, in Point p) pure nothrow @safe {

   static bool raySegI(in Point p, in Edge edge)

pure nothrow @safe { enum float epsilon = 0.001; // Increase error tolerance with (edge) { if (a.y > b.y) //swap(a, b); // if edge is mutable return raySegI(p, Edge(b, a)); if (p.y == a.y || p.y == b.y) //p.y += epsilon; // if p is mutable return raySegI(Point(p.x, p.y + epsilon), edge); if (p.y > b.y || p.y < a.y || p.x > max(a.x, b.x)) return false; if (p.x < min(a.x, b.x)) return true; immutable blue = (abs(a.x - p.x) > float.min_normal) ? ((p.y - a.y) / (p.x - a.x)) : float.max; immutable red = (abs(a.x - b.x) > float.min_normal) ? ((b.y - a.y) / (b.x - a.x)) : float.max; return blue >= red; } }

   return poly.edges.count!(e => raySegI(p, e)) % 2;

}

void main() {

   immutable Figure[] polys = [

{"Square", [ {{ 0.0, 0.0}, {10.0, 0.0}}, {{10.0, 0.0}, {10.0, 10.0}}, {{10.0, 10.0}, { 0.0, 10.0}}, {{ 0.0, 10.0}, { 0.0, 0.0}}]}, {"Square hole", [ {{ 0.0, 0.0}, {10.0, 0.0}}, {{10.0, 0.0}, {10.0, 10.0}}, {{10.0, 10.0}, { 0.0, 10.0}}, {{ 0.0, 10.0}, { 0.0, 0.0}}, {{ 2.5, 2.5}, { 7.5, 2.5}}, {{ 7.5, 2.5}, { 7.5, 7.5}}, {{ 7.5, 7.5}, { 2.5, 7.5}}, {{ 2.5, 7.5}, { 2.5, 2.5}}]}, {"Strange", [ {{ 0.0, 0.0}, { 2.5, 2.5}}, {{ 2.5, 2.5}, { 0.0, 10.0}}, {{ 0.0, 10.0}, { 2.5, 7.5}}, {{ 2.5, 7.5}, { 7.5, 7.5}}, {{ 7.5, 7.5}, {10.0, 10.0}}, {{10.0, 10.0}, {10.0, 0.0}}, {{10.0, 0}, { 2.5, 2.5}}]}, {"Exagon", [ {{ 3.0, 0.0}, { 7.0, 0.0}}, {{ 7.0, 0.0}, {10.0, 5.0}}, {{10.0, 5.0}, { 7.0, 10.0}}, {{ 7.0, 10.0}, { 3.0, 10.0}}, {{ 3.0, 10.0}, { 0.0, 5.0}}, {{ 0.0, 5.0}, { 3.0, 0.0}}]}, {"Strange2", [ // Our testing polygon {{-0.047600,3.619000},{-0.147595,3.618007}}, {{-0.147595,3.618007},{-0.111895,0.022807}}, {{-0.111895,0.022807},{-0.011900,0.023800}}, {{-0.011900,0.023800},{0.088095,0.024793}}, {{0.088095,0.024793},{0.052395,3.619993}}, {{0.052395,3.619993},{-0.047600,3.619000}}]} ];

   immutable Point[] testPoints = [{ 5, 5}, {5, 8}, {-10,  5}, {0, 5}, {10, 5}, {8, 5}, { 10, 10}, {-5.0,0.0238}];             // Last point need to be out of Strange2 polygon, but the result is true
   foreach (immutable poly; polys)

{

       writefln(`Is point inside figure "%s"?`, poly.name);
       foreach (immutable p; testPoints)
           writefln("  (%3s, %2s): %s", p.x, p.y, contains(poly, p));
       writeln;
   }

readln(); } </lang>

To see our polygon and test point you can use geogebra web :

 * go to http://www.geogebra.org/webstart/geogebra.html
 * enter in the input field : (-5.000000f,0.023800f)
 * enter in the input field : Polygon[(-0.047600,3.619000),(-0.147595,3.618007),(-0.111895,0.022807),(-0.011900,0.023800),(0.088095,0.024793),(0.052395,3.619993)]


We fix this issue by shifting the y coordinate of input point (adding epsilon) while it match with a vertex. Notice we are shifting the point before testing it against any segment, this increase the stability in the computation of the number of intersections.

Here is our working C++ implementation : <lang c++>

   template<class T, class Alloc>
   bool Polygon2<T, Alloc>::contains(const Vector2<T>& point, T epsilon) const
   {
       H3D_ASSERT(this->size() > 1);
       
       Vector2<T> currPt;
       // Insure point is not equal to one of the vertex
       Vector2<T> shiftedPoint(point);
       for (auto it = this->begin(); it != this->end();)
       {
           currPt = *it;
           
           if (math::epsilonEquals(currPt.y, shiftedPoint.y, epsilon))
           {
               shiftedPoint.y += epsilon;
               it = this->begin(); // Shift the point and recheck all the vertex (we could have shifted the point on a previous vertex)
           }
           else
           {
               ++it;
           }
       }
       
       int count = 0;
       Vector2<T> lastPt = this->back();
       for (auto it = this->begin(); it != this->end(); ++it)
       {
           currPt = *it;
           
           if (raySegI(shiftedPoint, lastPt, currPt, epsilon))
               ++count;
           
           lastPt = currPt;
       }
       return (bool)(count % 2);
   }
   template<class T, class Alloc>
   bool Polygon2<T, Alloc>::raySegI(const Vector2<T>& p, const Vector2<T>& a, const Vector2<T>& b, T epsilon) const
   {
       if (a.y > b.y)
           return raySegI(p, b, a, epsilon);
       H3D_ASSERT(math::epsilonEquals(p.y, a.y, epsilon) == false && math::epsilonEquals(p.y, b.y, epsilon) == false);
       if (p.y > b.y || p.y < a.y || p.x > std::max(a.x, b.x))
           return false;
       if (p.x < std::min(a.x, b.x))
           return true;
       T blue = (abs(a.x - p.x) > std::numeric_limits<T>::min()) ?
                    ((p.y - a.y) / (p.x - a.x)) :
                    std::numeric_limits<T>::max();
       T red = (abs(a.x - b.x) > std::numeric_limits<T>::min()) ?
                   ((b.y - a.y) / (b.x - a.x)) :
                   std::numeric_limits<T>::max();
       return blue >= red;
   }

</lang>

Elixir

point_in_polygon.ex: <lang Elixir>defmodule PointInPolygon do

 require Integer
 @doc """
 Check if point is inside a polygon
 ## Example
     iex> polygon = [[1, 2], [3, 4], [5, 2], [3, 0]]
     iex> point = [3, 2]
     iex> point_in_polygon?(polygon, point)
     true
 ## Example
     iex> polygon = [[1, 2], [3, 4], [5, 2], [3, 0]]
     iex> point = [1.5, 3]
     iex> point_in_polygon?(polygon, point)
     false
 """
 def point_in_polygon?(polygon, point) do
   polygon
   |> to_segments()
   |> Enum.reduce(0, fn segment, count ->
     apply(__MODULE__, :ray_intersects_segment, add_epsilon(segment, point)) + count
   end)
   |> Integer.is_odd()
 end
 def to_segments([p1 | _] = polygon) do
   polygon |> Enum.chunk_every(2, 1, [p1]) |> Enum.map(fn segment -> orient_segment(segment) end)
 end
 def orient_segment([a = [_ax, ay], b = [_bx, by]]) when by >= ay do
   [a, b]
 end
 def orient_segment([b, a]) do
   [a, b]
 end
 def add_epsilon(segment = [[_ax, ay], [_bx, by]], [px, py]) when py == ay or py == by do
   [segment, [px, py + 0.00000001]]
 end
 def add_epsilon(segment, point), do: [segment, point]
 def ray_intersects_segment([[_ax, ay], [_bx, by]], [_px, py]) when py < ay or py > by do
   0
 end
 # px >= max(ax, bx)
 def ray_intersects_segment([[ax, _ay], [bx, _by]], [px, _py])
     when (ax >= bx and px >= ax) or (bx >= ax and px >= bx) do
   0
 end
 # px < min(ax, bx)
 def ray_intersects_segment([[ax, _ay], [bx, _by]], [px, _py])
     when (ax <= bx and px < ax) or (bx <= ax and px < bx) do
   1
 end
 def ray_intersects_segment([[ax, ay], [bx, by]], [px, py]) do
   m_red = m_red(ax, ay, bx, by)
   m_blue = m_blue(ax, ay, px, py)
   case {m_blue, m_red} do
     {:infinity, _} ->
       1
     {_, :infinity} ->
       0
     {m_blue, m_red} when m_blue >= m_red ->
       1
     _ ->
       0
   end
 end
 def m_red(ax, ay, bx, by) when ax != bx do
   (by - ay) / (bx - ax)
 end
 def m_red(_, _, _, _) do
   :infinity
 end
 def m_blue(ax, ay, px, py) when ax != px do
   (py - ay) / (px - ax)
 end
 def m_blue(_, _, _, _) do
   :infinity
 end

end</lang>

Return to "Ray-casting algorithm" page.