Demo almost (?) working example: https://ellie-app.com/4h9F8FNcRPya1/1
For demo: Click to draw ray, and rotate camera with left and right to see ray. (As the origin is from the camera, you can't see it from the position it is created)
Context
I am working on an elm & elm-webgl project where I would like to know if the mouse is over an object when clicked. To do is I tried to implement a simple ray cast. What I need is two things:
1) The coordinate of the camera (This one is easy)
2) The coordinate/direction in 3D space of where was clicked
Problem
The steps to get from 2D view space to 3D world space as I understand are:
a) Make coordinates to be in a range of -1 to 1 relative to view port
b) Invert projection matrix and perspective matrix
c) Multiply projection and perspective matrix
d) Create Vector4 from normalised mouse coordinates
e) Multiply combined matrices with Vector4
f) Normalise result
Try so far
I have made a function to transform a Mouse.Position to a coordinate to draw a line to:
getClickPosition : Model -> Mouse.Position -> Vec3
getClickPosition model pos =
let
x =
toFloat pos.x
y =
toFloat pos.y
normalizedPosition =
( (x * 2) / 1000 - 1, (1 - y / 1000 * 2) )
homogeneousClipCoordinates =
Vec4.vec4
(Tuple.first normalizedPosition)
(Tuple.second normalizedPosition)
-1
1
inversedProjectionMatrix =
Maybe.withDefault Mat4.identity (Mat4.inverse (camera model))
inversedPerspectiveMatrix =
Maybe.withDefault Mat4.identity (Mat4.inverse perspective)
inversedMatrix2 =
Mat4.mul inversedProjectionMatrix inversedPerspectiveMatrix
to =
Vec4.vec4
(Tuple.first normalizedPosition)
(Tuple.second normalizedPosition)
1
1
toInversed =
mulVector inversedMatrix2 to
toNorm =
Vec4.normalize toInversed
toVec3 =
vec3 (Vec4.getX toNorm) (Vec4.getY toNorm) (Vec4.getZ toNorm)
in
toVec3
Result
The result of this function is that the rays are too much to the center to where I click. I added a screenshot where I clicked in all four of the top face of the cube. If I click on the center of the viewport the ray will be correctly positioned.
It feels close, but not quite there yet and I can't figure out what I am doing wrong!
After trying other approaches I found a solution:
getClickPosition : Model -> Mouse.Position -> Vec3
getClickPosition model pos =
let
x =
toFloat pos.x
y =
toFloat pos.y
normalizedPosition =
( (x * 2) / 1000 - 1, (1 - y / 1000 * 2) )
homogeneousClipCoordinates =
Vec4.vec4
(Tuple.first normalizedPosition)
(Tuple.second normalizedPosition)
-1
1
inversedViewMatrix =
Maybe.withDefault Mat4.identity (Mat4.inverse (camera model))
inversedProjectionMatrix =
Maybe.withDefault Mat4.identity (Mat4.inverse perspective)
vec4CameraCoordinates = mulVector inversedProjectionMatrix homogeneousClipCoordinates
direction = Vec4.vec4 (Vec4.getX vec4CameraCoordinates) (Vec4.getY vec4CameraCoordinates) -1 0
vec4WorldCoordinates = mulVector inversedViewMatrix direction
vec3WorldCoordinates = vec3 (Vec4.getX vec4WorldCoordinates) (Vec4.getY vec4WorldCoordinates) (Vec4.getZ vec4WorldCoordinates)
normalizedVec3WorldCoordinates = Vec3.normalize vec3WorldCoordinates
origin = model.cameraPos
scaledDirection = Vec3.scale 20 normalizedVec3WorldCoordinates
destination = Vec3.add origin scaledDirection
in
destination
I left it as verbose as possible, if someone finds I use incorrect terminology please make a comment and I will update the answer.
I am sure there are lots of optimisations possible (Multiplying matrices before inverting or combining some of the steps.)
Updated the ellie app here: https://ellie-app.com/4hZ9s8S92PSa1/0
I understand that Scale Layer works as z = x * y where y is "broadcasted" to match the shape of x.
Here is my question:
I have bottom[0] size as N*C*H*W, with bottom[1] as N*1*H*W,
and I want to tile bottom[1] to N*C*H*W, then computing the element-wise product.
How should I set the parameters, especially "axis"?
Thx a lot.
Based on scale_layer.cpp axis is the starting axis for scaling:
axis_ = bottom[0]->CanonicalAxisIndex(param.axis());
...
const vector<int>::const_iterator& shape_start =
bottom[0]->shape().begin() + axis_;
const vector<int>::const_iterator& shape_end =
(num_axes == -1) ? bottom[0]->shape().end() : (shape_start + num_axes);
vector<int> scale_shape(shape_start, shape_end);
So, in your case you can set axis parameter to either 0 or 1.
There is also another parameter num_axis, which from my understanding, is the number of axes that are scaled. In your case, it can be set to 2 or 1 depending on you axis parameter. (axis=1, num_axis=1 or axis =0, num_axis=2).
I have never used this layer, and these conclusions are only based on the source code of scale_layer.
I am not hundred percent sure. But you could try two things. In concat layer, axis = 0, means the first dimension(N) in your N * C * H * W, and axis = 1, means second dimension (C). so you could try with axis = 1, and scale parameter of 3. I am not completely sure. But you could try and let me know what you get.
Another thing you can try is to use Concat layer with axis = 1.
Recently I had the idea to make a pendulum out of points using Processing, and with a little learning I solved it easily:
int contador = 0;
int curvatura = 2;
float pendulo;
void setup(){
size(300,300);
}
void draw(){
background(100);
contador = (contador + 1) % 360; //"CONTADOR" GOES FROM 0 TO 359
pendulo = sin(radians(contador))*curvatura; //"PENDULO" EQUALS THE SIN OF CONTADOR, SO IT GOES FROM 1 TO -1 REPEATEDLY, THEN IS MULTIPLIED TO EMPHASIZE OR REDUCE THE CURVATURE OF THE LINE.
tallo(width/2,height/3);
println(pendulo);
}
void tallo (int x, int y){ //THE FUNTION TO DRAW THE DOTTED LINE
pushMatrix();
translate(x,y);
float _y = 0.0;
for(int i = 0; i < 25; i++){ //CREATES THE POINTS SEQUENCE.
ellipse(0,0,5,5);
_y+=5;
rotate(radians(pendulo)); //ROTATE THEM ON EACH ITERATION, THIS MAKES THE SPIRAL.
}
popMatrix();
}
So, in a brief, what I did was a function that changed every point position with the rotate fuction, and then I just had to draw the ellipses in the origin coordinates as that is the real thing that changes position and creates the pendulum ilussion.
[capture example, I just need 2 more points if you are so gentile :)]
[capture example]
[capture example]
Everything was OK that far. The problem appeared when I tried to replace the ellipses for a path made of vertices. The problem is obvious: the path is never (visually) made because all vertices would be 0,0 as they move along with the zero coordinates.
So, in order to make the path possible, I need the absolute values for each vertex; and there's the question: How do I get them?
What I know I have to do is to remove the transform functions, create the variables for the X and Y position and update them inside the for, but then what? That's why I cleared this is a maths issue, which operation I have to add in the X and Y variables in order to make the path and its curvature possible?
void tallo (int x, int y){
pushMatrix();
translate(x,y);
//NOW WE START WITH THE CHANGES. LET'S DECLARE THE VARIABLES FOR THE COORDINATES
float _x = 0.0;
float _y = 0.0;
beginShape();
for(int i = 0; i < 25; i++){ //CREATES THE DOTS.
vertex(_x,_y); //CHANGING TO VERTICES AND CALLING THE NEW VARIABLES, OK.
//rotate(radians(pendulo)); <--- HERE IS MY PROBLEM. HOW DO I CONVERT THIS INTO X AND Y COORDINATES?
//_x = _x + ????;
_y = _y + 5 /* + ???? */;
}
endShape();
popMatrix();
}
We need to have in mind that pendulo's x and y values changes in each iteration of the for, it doesn't has to add the same quantity each time. The addition must be progressive. Otherwise, we would see a straight line rotating instead of a curve accentuating its curvature (if you increase curvatura's value to a number greater than 20, you will notice the spiral)
So, rotating the coordinates was a great solution to it, now it's kind of a muddle to think the mathematical solution to the x and y coordinates for the spiral, my secondary's knowledges aren't enough. I know I have to create another variable inside the for in order to do this progression, but what operation should it have?
I would be really glad to know, maths
You could use simple trigonometry. You know the angle and the hypotenuse, so you use cos to get the relative x position, and sin to the y. The position would be relative to the central point.
But before i explain in detail and draw some explanations, let me propose another solution: PVectors
void setup() {
size(400,400);
frameRate(60);
center = new PVector(width/2, height/3); //defined here because width and height only are set after size()
}
void draw() {
background(255);
fill(0);
stroke(0);
angle = arc_magn*sin( (float) frameCount/60 );
draw_pendulum( center );
}
PVector center;
float angle = 0;
float arc_magn = HALF_PI;
float wire_length = 150;
float rotation_angle = PI/20 /60 ; //we divide it by 60 so the first part is the rotation in one second
void draw_pendulum(PVector origin){
PVector temp_vect = PVector.fromAngle( angle + HALF_PI);
temp_vect.setMag(wire_length);
PVector final_pos = new PVector(origin.x+temp_vect.x, origin.y+temp_vect.y );
ellipse( final_pos.x, final_pos.y, 40, 40);
line(origin.x, origin.y, final_pos.x, final_pos.y);
}
You use PVector class static method fromAngle( float angle ) that returns a unity vector of the given angle, then use .setMag() to define it's length.
Those PVector methods will take care of the trigonometry for you.
If you still want to know the math behind it, i can make another example.
I have the last two CGPoints from a Array which contains points of line drawn by the user . i need to extend the line upto a fixed distance at the same angle. so i first calculate the angle between the last two points with the help of following code
-(CGFloat)angleBetweenFirstPoint:(CGPoint)firstPoint ToSecondPoint:(CGPoint)secondPoint
{
CGPoint diff = ccpSub(secondPoint, firstPoint);
NSLog(#"difference point %f , %f",diff.x,diff.y);
CGFloat res = atan2(diff.y, diff.x);
/*if ( res < 0 )
{
res = (0.5 * M_PI) + res;
}
if ( dx<0 && dy>0 ) { // 2nd quadrant
res += 0.5 * M_PI;
} else if ( dx<0 && dy<0 ) { // 3rd quadrant
res += M_PI;
} else if ( dx>0 && dy<0 ) { // 4th quadrant
res += M_PI + (0.5 * M_PI);
}*/
//res=res*180/M_PI;
res = CC_RADIANS_TO_DEGREES(res);
return res;
}
After calculating the angle i find the extend point with the help of following maths
-(void)extendLine
{
lineAngle = [self angleBetweenFirstPoint:pointD ToSecondPoint:endPt];
extendEndPt.x = endPt.x - cos(lineAngle) * 200;
extendEndPt.y = endPt.y - sin(lineAngle) * 200;
// draw line unto extended point
}
But the point i am getting is not right to draw the extended line at the same angle as the original line.
I think it is because i am not getting the right angle between those last points.. what am i possibly doing wrong?? Do i need to consider the whole quadrant system while considering the angle and how? and m working in landscape mode. does that make any difference??
Ye gods, you are doing this in a way that is WILDLY INCREDIBLY over-complicated.
Skip all of the crapola with angles. You don't need it. Period. Do it all with vectors, and very simple ones. First of all, I'll assume that you are given two points, P1 and P2. You wish to find a new point P3, that is a known distance (d) from P2, along the line that connects the two points.
All you need do is first, compute a vector that points along the line in question.
V = P2 - P1;
I've written it as if I am writing in MATLAB, but all this means is to subtract the x and y coordinates of the two points.
Next, scale the vector V to have unit length.
V = V/sqrt(V(1)^2 + V(2)^2);
Dividing the components of the vector V by the length (or 2-norm if you prefer) of that vector creates a vector with unit norm. That norm is just the square root of the sum of squares of the elements of V, so it is clearly the length of the vector.
Now it is simple to compute P3.
P3 = P2 + d*V;
P3 will lie at a distance of d units from P2, in the direction of the line away from point P1. Nothing sophisticated required. No angles computed. No worry about quadrants.
Learn to use vectors. They are your friends, or at the least, they can be if you let them.
Given two rectangles with x, y, width, height in pixels and a rotation value in degrees -- how do I calculate the closest distance of their outlines toward each other?
Background: In a game written in Lua I'm randomly generating maps, but want to ensure certain rectangles aren't too close to each other -- this is needed because maps become unsolvable if the rectangles get into certain close-distance position, as a ball needs to pass between them. Speed isn't a huge issue as I don't have many rectangles and the map is just generated once per level. Previous links I found on StackOverflow are this and this
Many thanks in advance!
Not in Lua, a Python code based on M Katz's suggestion:
def rect_distance((x1, y1, x1b, y1b), (x2, y2, x2b, y2b)):
left = x2b < x1
right = x1b < x2
bottom = y2b < y1
top = y1b < y2
if top and left:
return dist((x1, y1b), (x2b, y2))
elif left and bottom:
return dist((x1, y1), (x2b, y2b))
elif bottom and right:
return dist((x1b, y1), (x2, y2b))
elif right and top:
return dist((x1b, y1b), (x2, y2))
elif left:
return x1 - x2b
elif right:
return x2 - x1b
elif bottom:
return y1 - y2b
elif top:
return y2 - y1b
else: # rectangles intersect
return 0.
where
dist is the euclidean distance between points
rect. 1 is formed by points (x1, y1) and (x1b, y1b)
rect. 2 is formed by points (x2, y2) and (x2b, y2b)
Edit: As OK points out, this solution assumes all the rectangles are upright. To make it work for rotated rectangles as the OP asks you'd also have to compute the distance from the corners of each rectangle to the closest side of the other rectangle. But you can avoid doing that computation in most cases if the point is above or below both end points of the line segment, and to the left or right of both line segments (in telephone positions 1, 3, 7, or 9 with respect to the line segment).
Agnius's answer relies on a DistanceBetweenLineSegments() function. Here is a case analysis that does not:
(1) Check if the rects intersect. If so, the distance between them is 0.
(2) If not, think of r2 as the center of a telephone key pad, #5.
(3) r1 may be fully in one of the extreme quadrants (#1, #3, #7, or #9). If so
the distance is the distance from one rect corner to another (e.g., if r1 is
in quadrant #1, the distance is the distance from the lower-right corner of
r1 to the upper-left corner of r2).
(4) Otherwise r1 is to the left, right, above, or below r2 and the distance is
the distance between the relevant sides (e.g., if r1 is above, the distance
is the distance between r1's low y and r2's high y).
Actually there is a fast mathematical solution.
Length(Max((0, 0), Abs(Center - otherCenter) - (Extent + otherExtent)))
Where Center = ((Maximum - Minimum) / 2) + Minimum and Extent = (Maximum - Minimum) / 2.
Basically the code above zero's axis which are overlapping and therefore the distance is always correct.
It's preferable to keep the rectangle in this format as it's preferable in many situations ( a.e. rotations are much easier ).
Pseudo-code:
distance_between_rectangles = some_scary_big_number;
For each edge1 in Rectangle1:
For each edge2 in Rectangle2:
distance = calculate shortest distance between edge1 and edge2
if (distance < distance_between_rectangles)
distance_between_rectangles = distance
There are many algorithms to solve this and Agnius algorithm works fine. However I prefer the below since it seems more intuitive (you can do it on a piece of paper) and they don't rely on finding the smallest distance between lines but rather the distance between a point and a line.
The hard part is implementing the mathematical functions to find the distance between a line and a point, and to find if a point is facing a line. You can solve all this with simple trigonometry though. I have below the methodologies to do this.
For polygons (triangles, rectangles, hexagons, etc.) in arbitrary angles
If polygons overlap, return 0
Draw a line between the centres of the two polygons.
Choose the intersecting edge from each polygon. (Here we reduce the problem)
Find the smallest distance from these two edges. (You could just loop through each 4 points and look for the smallest distance to the edge of the other shape).
These algorithms work as long as any two edges of the shape don't create angles more than 180 degrees. The reason is that if something is above 180 degrees then it means that the some corners are inflated inside, like in a star.
Smallest distance between an edge and a point
If point is not facing the face, then return the smallest of the two distances between the point and the edge cornerns.
Draw a triangle from the three points (edge's points plus the solo point).
We can easily get the distances between the three drawn lines with Pythagorean Theorem.
Get the area of the triangle with Heron's formula.
Calculate the height now with Area = 12⋅base⋅height with base being the edge's length.
Check to see if a point faces an edge
As before you make a triangle from an edge and a point. Now using the Cosine law you can find all the angles with just knowing the edge distances. As long as each angle from the edge to the point is below 90 degrees, the point is facing the edge.
I have an implementation in Python for all this here if you are interested.
This question depends on what kind of distance. Do you want, distance of centers, distance of edges or distance of closest corners?
I assume you mean the last one. If the X and Y values indicate the center of the rectangle then you can find each the corners by applying this trick
//Pseudo code
Vector2 BottomLeftCorner = new Vector2(width / 2, heigth / 2);
BottomLeftCorner = BottomLeftCorner * Matrix.CreateRotation(MathHelper.ToRadians(degrees));
//If LUA has no built in Vector/Matrix calculus search for "rotate Vector" on the web.
//this helps: http://www.kirupa.com/forum/archive/index.php/t-12181.html
BottomLeftCorner += new Vector2(X, Y); //add the origin so that we have to world position.
Do this for all corners of all rectangles, then just loop over all corners and calculate the distance (just abs(v1 - v2)).
I hope this helps you
I just wrote the code for that in n-dimensions. I couldn't find a general solution easily.
// considering a rectangle object that contains two points (min and max)
double distance(const rectangle& a, const rectangle& b) const {
// whatever type you are using for points
point_type closest_point;
for (size_t i = 0; i < b.dimensions(); ++i) {
closest_point[i] = b.min[i] > a.min[i] ? a.max[i] : a.min[i];
}
// use usual euclidian distance here
return distance(a, closest_point);
}
For calculating the distance between a rectangle and a point you can:
double distance(const rectangle& a, const point_type& p) const {
double dist = 0.0;
for (size_t i = 0; i < dimensions(); ++i) {
double di = std::max(std::max(a.min[i] - p[i], p[i] - a.max[i]), 0.0);
dist += di * di;
}
return sqrt(dist);
}
If you want to rotate one of the rectangles, you need to rotate the coordinate system.
If you want to rotate both rectangles, you can rotate the coordinate system for rectangle a. Then we have to change this line:
closest_point[i] = b.min[i] > a.min[i] ? a.max[i] : a.min[i];
because this considers there is only one candidate as the closest vertex in b. You have to change it to check the distance to all vertexes in b. It's always one of the vertexes.
See: https://i.stack.imgur.com/EKJmr.png
My approach to solving the problem:
Combine the two rectangles into one large rectangle
Subtract from the large rectangle the first rectangle and the second
rectangle
What is left after the subtraction is a rectangle between the two
rectangles, the diagonal of this rectangle is the distance between
the two rectangles.
Here is an example in C#
public static double GetRectDistance(this System.Drawing.Rectangle rect1, System.Drawing.Rectangle rect2)
{
if (rect1.IntersectsWith(rect2))
{
return 0;
}
var rectUnion = System.Drawing.Rectangle.Union(rect1, rect2);
rectUnion.Width -= rect1.Width + rect2.Width;
rectUnion.Width = Math.Max(0, rectUnion.Width);
rectUnion.Height -= rect1.Height + rect2.Height;
rectUnion.Height = Math.Max(0, rectUnion.Height);
return rectUnion.Diagonal();
}
public static double Diagonal(this System.Drawing.Rectangle rect)
{
return Math.Sqrt(rect.Height * rect.Height + rect.Width * rect.Width);
}
Please check this for Java, it has the constraint all rectangles are parallel, it returns 0 for all intersecting rectangles:
public static double findClosest(Rectangle rec1, Rectangle rec2) {
double x1, x2, y1, y2;
double w, h;
if (rec1.x > rec2.x) {
x1 = rec2.x; w = rec2.width; x2 = rec1.x;
} else {
x1 = rec1.x; w = rec1.width; x2 = rec2.x;
}
if (rec1.y > rec2.y) {
y1 = rec2.y; h = rec2.height; y2 = rec1.y;
} else {
y1 = rec1.y; h = rec1.height; y2 = rec2.y;
}
double a = Math.max(0, x2 - x1 - w);
double b = Math.max(0, y2 - y1 - h);
return Math.sqrt(a*a+b*b);
}
Another solution, which calculates a number of points on the rectangle and choses the pair with the smallest distance.
Pros: works for all polygons.
Cons: a little bit less accurate and slower.
import numpy as np
import math
POINTS_PER_LINE = 100
# get points on polygon outer lines
# format of polygons: ((x1, y1), (x2, y2), ...)
def get_points_on_polygon(poly, points_per_line=POINTS_PER_LINE):
all_res = []
for i in range(len(poly)):
a = poly[i]
if i == 0:
b = poly[-1]
else:
b = poly[i-1]
res = list(np.linspace(a, b, points_per_line))
all_res += res
return all_res
# compute minimum distance between two polygons
# format of polygons: ((x1, y1), (x2, y2), ...)
def min_poly_distance(poly1, poly2, points_per_line=POINTS_PER_LINE):
poly1_points = get_points_on_polygon(poly1, points_per_line=points_per_line)
poly2_points = get_points_on_polygon(poly2, points_per_line=points_per_line)
distance = min([math.sqrt((a[0] - b[0])**2 + (a[1] - b[1])**2) for a in poly1_points for b in poly2_points])
# slower
# distance = min([np.linalg.norm(a - b) for a in poly1_points for b in poly2_points])
return distance