3D Projection Modification - Encode Z/W into Z - math

This is a little tricky to explain, so bare with me. I'm attempting to design a 2D projection matrix that takes 2D pixel coordinates along with a custom world-space depth value, and converts to clip-space.
The idea is that it would allow drawing elements based on screen coordinates, but at specific depths, so that these elements would interact on the depth buffer with normal 3D elements. However, I want x and y coordinates to remain the same scale at every depth. I only want depth to influence the depth buffer, and not coordinates or scale.
After the vertex shader, the GPU sets depth_buffer=z/w. However, it also scales x/w and y/w, which creates the depth scaling I want to avoid. This means I must make sure my final clip-space w coordinate ends up being 1.0, to avoid those things. I think I could also adopt to scale x and y by w, to cancel out the divide, but I would rather do the former, if possible.
This is the process that my 3D projection matrix uses to convert depth into clip space (d = depth, n = near distance, f = far distance)
z = f/(f-n) * d + f/(f-n) * -n;
w = d;
This is how I would like to setup my 2D projection matrix. Compared to the 3D version, it would divide both attributes by the input depth. This would simulate having z/w encoded into just the z value.
z = ( f/(f-n) * d + f/(f-n) * -n ) / d;
w = d / d;
I think this turns into something like..
r = f/(f-n); // for less crazy math
z = r + ( r * -n ) / d;
w = 1.0;
However, I can't seem to wrap my math around the values that I would need to plug into my matrix to get this result. It looks like I would need to set my matrix up to perform a division by depth. Is that even possible? Can anyone help me figure out the values I need to plug into my matrix at m[2][2] and m[3][2] (m._33 and m._43) to make something like this happen?
Note my 3D projection matrix uses the following properties to generate the final z value:
m._33 = f / (f-n); // depth scale
m._43 = -(f / (f-n)) * n; // depth offset
Edit: After thinking about this a little more, I realized that the rate of change of the depth buffer is not linear, and I'm pretty sure a matrix can only perform linear change when its input is linear. If that is the case, then what I'm trying to do wouldn't be possible. However, I'm still open to any ideas that are in the same ball park, if anyone has one. I know that I can get what I want by simply doing pos.z /= pos.w; pos.w = 1; in the vertex shader, but I was really hoping to make it all happen in the projection matrix, if possible.

In case anyone is attempting to do this, it cannot be done. Without black magic, there is apparently no way to divide values with a matrix, unless of course the diviser is a constant or etc, where you can swap out a scaler with 1/x. I resorted to performing the operation in the shader in the end.

Related

3D bounding box for an item with three axis rotations

I'm trying to find the <x,y,z> size of what would end up being a bounding box for a rotated shape in all three axis rotations. Though to help keep it simple, the example demonstrated below has only the x axis rotated.
vector Size = <10,1,0.5>; vector Deg = <22.5,0,0>
if(Deg.x > 0 && Deg.y == 0 && Deg.z == 0){
Y1 = Cos(Deg.x) * Size.y + Sin(Deg.x) * Size.z;
Z1 = Cos(Deg.x) * Size.z + Sin(Deg.x) * Size.y;}
Below are for the y and z rotations, that is if you decided to change the degrees to say <0,22.5,0> and <0,0,22.5>.
if(Deg.y > 0 && Deg.x == 0 && Deg.z == 0){
X2 = Cos(Deg.y) * Size.x + Sin(Deg.y) * Size.z;
Z2 = Cos(Deg.y) * Size.z + Sin(Deg.y) * Size.x;}
if(Deg.z > 0 && Deg.x == 0 && Deg.y == 0){
X3 = Cos(Deg.z) * Size.x + Sin(Deg.z) * Size.y;
Y3 = Cos(Deg.z) * Size.y + Sin(Deg.z) * Size.x;}
Though the part I get lost at is, where do I go from here if when you have the rotation in two or three axis. Such as <22.5,22.5,0> or <22.5,22.5.22.5>
Is there a website with a tutorial or example equations I could review or are there any hints or ideas of what I could do to figure this out.
EDIT:
I do want to add that the comment from Nico helped, in that what I'm asking about is called: Axis-Aligned Bounding Box or AABB for short.
As for JohanC comment about the 22.5, yes the Deg = Degree. Also yes you'll have to turn the Degree into Radians, but I put it as Degree in the example for the Sin and Cos input to keep it simple.
If you're wondering how this question might be useful. Well as an example you'd need to know the AABB to help in an equation to keep the item in question flush with the surface its sitting on if you were to say resize the item while it had <x,y,z> rotations that weren't at perfect zero rotations.
I found my own solution after a lot of testing and debugging. I'll show and explain a small portion of the script I created since functions and options from coding language to language may vary.
A few details about the example script below. Deg = Degrees. Though yes what JohanC mentioned about radians is correct, since even the language I'm using I have to convert degrees to radians with a function. Not every language or calculator is like that, yet for intents and purposes to make it easier to read, I did take out the excess while keeping the meat of the code/equation intact so to speak. Also in this example, I'm rotating it in the Z axis to demonstrate and as a starting point.
vector Size = <10,1,0.5>; //Insert your own coding language function
//for splitting the size in half with positive and negative halves.
vector min = <-5,-0.5,-0.25>; vector max = <5,0.5,0.25>;
//You'll need at least 8 points in total to do this correctly.
//I already have 8 points added down below, labeled p1 - p8
vector p1 = <max.x,max.y,max.z>; vector p2 = <max.x,min.y,max.z>;
vector p3 = <max.x,max.y,min.z>; vector p4 = <max.x,min.y,min.z>;
vector p5 = <min.x,max.y,max.z>; vector p6 = <min.x,min.y,max.z>;
vector p7 = <min.x,max.y,min.z>; vector p8 = <min.x,min.y,min.z>;
vector Deg = <0,0,22.5>
//This will give you an idea of how to setup the x,y,z for each point
//equation to make / change to fit the rotations. Plus to make it compact
//I suggest that you use list, a while loop, and etc.
x1 = p1.x * llCos(Deg.z) - p1.y * llSin(Deg.z);
y1 = p1.y * llCos(Deg.z) + p1.x * llSin(Deg.z);
A lot of what I kept out of the example above is several list, true and false statements, a while loop, and few other things. Though it's kept simple to show the less complicated portion of it and the fact that not every language has the same functions available.
Though once when you get all of the location data of the 8 points collected, you can then put each of the x, y, z point information into its own list. Then run the equation (edited as needed for each rotation). After that, get the max and min from each axis rotation output. Then take the maximum and minus the minimum from it since the minimum will always be a negative. That right there will give you the <x,y,z> size of your bounding box.
I do want to add that the link from Nico helped slightly, though the only flaw in it is "if" the language you're using allows a matrix or "if" you're able to create a multidimensional array. If you can manage that, then use the link Nico gave to see if it helps.
Also some of JohanC tips/hints helped as well. Speaking of which, the part about "using the output of the vector from one rotation to the next" works well if and only if you're using the 8 points method. Otherwise if you try it with the bounding box size of one to the next, the first rotation to the second will work fine because its only moving in 2D at that point, but by the time you try to make the third rotation with the second rotations bounding box it won't work because it'll be going from 2D to 3D.
Note - If you think you have a better way to do the equation or simpler way to explain it, feel free to add in your own answer.

What's a simple way of warping an image with a given set of points?

I'd like to implement image morphing, for which I need to be able to deform the image with given set of points and their destination positions (where they will be "dragged"). I am looking for a simple and easy solution that gets the job done, it doesn't have to look great or be extremely fast.
This is an example what I need:
Let's say I have an image and a set of only one deforming point [0.5,0.5] which will have its destination at [0.6,0.5] (or we can say its movement vector is [0.1,0.0]). This means I want to move the very center pixel of the image by 0.1 to the right. Neighboring pixels in some given radius r need to of course be "dragged along" a little with this pixel.
My idea was to do it like this:
I'll make a function mapping the source image positions to destination positions depending on the deformation point set provided.
I will then have to find the inverse function of this function, because I have to perform the transformation by going through destination pixels and seeing "where the point had to come from to come to this position".
My function from step 1 looked like this:
p2 = p1 + ( 1 / ( (distance(p1,p0) / r)^2 + 1 ) ) * s
where
p0 ([x,y] vector) is the deformation point position.
p1 ([x,y] vector) is any given point in the source image.
p2 ([x,y] vector) is the position, to where p1 will be moved.
s ([x,y] vector) is movement vector of deformation point and says in which direction and how far p0 will be dragged.
r (scalar) is the radius, just some number.
I have problem with step number 2. The calculation of the inverse function seems a little too complex to me and so I wonder:
If there is an easy solution for finding the inverse function, or
if there is a better function for which finding the inverse function is simple, or
if there is an entirely different way of doing all this that is simple?
Here's the solution in Python - I did what Yves Daoust recommended and simply tried to use the forward function as the inverse function (switching the source and destination). I also altered the function slightly, changing exponents and other values produces different results. Here's the code:
from PIL import Image
import math
def vector_length(vector):
return math.sqrt(vector[0] ** 2 + vector[1] ** 2)
def points_distance(point1, point2):
return vector_length((point1[0] - point2[0],point1[1] - point2[1]))
def clamp(value, minimum, maximum):
return max(min(value,maximum),minimum)
## Warps an image accoording to given points and shift vectors.
#
# #param image input image
# #param points list of (x, y, dx, dy) tuples
# #return warped image
def warp(image, points):
result = img = Image.new("RGB",image.size,"black")
image_pixels = image.load()
result_pixels = result.load()
for y in range(image.size[1]):
for x in range(image.size[0]):
offset = [0,0]
for point in points:
point_position = (point[0] + point[2],point[1] + point[3])
shift_vector = (point[2],point[3])
helper = 1.0 / (3 * (points_distance((x,y),point_position) / vector_length(shift_vector)) ** 4 + 1)
offset[0] -= helper * shift_vector[0]
offset[1] -= helper * shift_vector[1]
coords = (clamp(x + int(offset[0]),0,image.size[0] - 1),clamp(y + int(offset[1]),0,image.size[1] - 1))
result_pixels[x,y] = image_pixels[coords[0],coords[1]]
return result
image = Image.open("test.png")
image = warp(image,[(210,296,100,0), (101,97,-30,-10), (77,473,50,-100)])
image.save("output.png","PNG")
You don't need to construct the direct function and invert it. Directly compute the inverse function, by swapping the roles of the source and destination points.
You need some form of bivariate interpolation, have a look at radial basis function interpolation. It requires to solve a linear system of equations.
Inverse distance weighting (similar to your proposal) is the easiest to implement but I am afraid it will give disappointing results.
https://en.wikipedia.org/wiki/Multivariate_interpolation#Irregular_grid_.28scattered_data.29

Arrange X amount of things evenly around a point in 3d space

If I have X amount of things (lets just randomly say 300)
Is there an algorithm that will arrange these things somewhat evenly around a central point? Like a 100 sided dice or a 3d mesh of a sphere?
Id rather have the things somewhat evenly spaced like this..
Rather than this polar way..
ps. For those interested, wondering why do I want to do this?
Well I'm doing these for fun, and after completing #7 I decided I'd like to represent the array of wires in 3d in Unity and watch them operate in a slowed down manner.
Here is a simple transformation that maps a uniform sample in the rectangle [0, 2 pi] x [-1, 1] onto a uniform sample on the sphere of radius r:
T(phi, z) = (r cos(phi) sqrt(1 - z^2), r sin(phi) sqrt(1 - zˆ2), r z)
The reason why this transformation produces uniform samples on the sphere is that the area of any region T(U) obtained by transforming the region U from the rectangle does not depend on U but on the area of U.
To prove this mathematically it is enough to verify that the norm of the vectorial product
| ∂T/∂phi x ∂T/∂z |
is constant (the area on the sphere is the integral of this vectorial product w.r.t. phi and z).
Summarizing
To produce a random sample uniformly distributed in the Sphere of radius r do the following:
Produce a random sample (phi_1, ..., phi_n) uniformly distributed in [0, 2 pi].
Produce a random sample (z_1, ..., z_n) uniformly distributed in [-1, 1].
For every pair (phi_j, z_k) calculate T(phi_j, z_k) using the formula above.
Here's a three-step approach. 1a) Make more points than you need. 1b) Remove some. 2) Adjust the rest.
1a) To make more points that you need, take any quasiregular polyhedron with sides that tessellate (triangles, squares, diamonds). Tesselate the spherical faces by subdivision, generating more vertices. For example, if you use the regular icosahedron you get geodesic domes. (Subdivide by 2, you get the dual to the C60 buckyball.) Working out exact formulas isn't hard. The number of new vertices per face is quadratic in the subdivision.
1b) Randomly remove enough points to get you down to your target number.
2) Use a force-directed layout algorithm to redistribute the vertices over the sphere. The underlying force graph is just that provided by the nearest neighbors in your underlying tesselation.
There are other ways to do step 1), such as just generating random points in any distribution. There is an advantage of starting with a quasiregular figure, though. Force-directed algorithms have a reputation for poor convergence in some cases. By starting with something that's already mostly optimal, you'll bypass most all of any convergence problems you might have.
One elegant solution I came across recently is a spherical fibonacci lattice (http://extremelearning.com.au/how-to-evenly-distribute-points-on-a-sphere-more-effectively-than-the-canonical-fibonacci-lattice/)
The nice thing about it is that you can specify the exact number of points you want
// C# Code example
Vector3[] SphericalFibonacciLattice(int n) {
Vector3[] res = new Vector3[n];
float goldenRatio = (1.0f + MathF.Sqrt(5.0f)) * 0.5f;
for(int i = 0; i < n; i++)
{
float theta = 2.0f * MathF.PI * i / goldenRatio;
float phi = MathF.Acos(1.0f - 2.0f * (i + 0.5f) / n);
Vector3 p = new Vector3(MathF.Cos(theta) * MathF.Sin(phi),
MathF.Sin(theta) * MathF.Sin(phi),
MathF.Cos(phi));
res[i] = p;
}
return res;
}
The linked article extends on this to create an even more uniform distribution, but even this basic version creates very nice results.

Algorithm to find all points on a 2D grid some distance away from another point

I have some point on a 2D grid (x, y) and I need to find all points that are n distance away from that point. The way I'm measuring distance is by using the distance formula between the two points. Anyone know how to do this?
Edit: Just for reference, what I'm trying to do is to write some AI path finding that will maintain some distance away from a target in a system that uses grid based locations. Currently I'm using A* path finding, but I'm not sure if that matters or makes a difference since I'm kind of new to this stuff.
Here's what I would do:
First filter out all points that are further than D on either x or y. These are certainly outside the circle of radius D. This is a much simpler computation, and it can quickly eliminate a lot of work. This is a outer bounding-box optimization.
You can also use an inner bounding-box optimization. If the points are closer than D * sqrt(2)/2 on either x or y, then they're certainly within the circle of radius D. This is also cheaper than calculating the distance formula.
Then you have a smaller number of candidate points that may be within the circle of radius D. For these, use the distance formula. Remember that if D = sqrt(Δx2+Δy2), then D2 = Δx2+Δy2.
So you can skip the cost of calculating square root.
So in pseudocode, you could do the following:
for each point
begin
if test 1 indicates the point is outside the outer bounding box,
then skip this point
if test 2 indicates the point is inside the inner bounding box,
then keep this point
if test 3 indicates the point is inside the radius of the circle,
then keep this point
end
This problem is known as range query. The brute force solution is just as you described: computed the distance of all points from the reference point and return those whose distance is less than the desired range value.
The brute force algorithm is O(N^2). There are, however, more efficient algorithms that employ spatial indexes to reduce algorithm complexity and the number of distance calculations. For example, you can use a R-Tree to index your points.
Its called nearest neighbor search. More at http://en.wikipedia.org/wiki/Nearest_neighbor_search
There are open libraries for that. I have used one written for C and recommend it: http://www.cs.umd.edu/~mount/ANN/. ANN stands for Approximate Nearest Neighbor, however, you can turn the approximation off and find the exact nearest neighbors.
This wouldn't use the distance formula, but if you're looking for points exactly n distance away, perhaps you could use sin/cos?
In pseudocode:
for degrees in range(360):
x = cos(degrees) * n
y = sin(degrees) * n
print x, y
That would print every point n away in 360 degree increments.
Java implementation:
public static Set<Point> findNearbyPoints(Set<Point> pts, Point centerPt, double radius) {
Set<Point> nearbyPtsSet = new HashSet<Point>();
double innerBound = radius * (Math.sqrt(2.0) / 2.0);
double radiusSq = radius * radius;
for (Point pt : pts) {
double xDist = Math.abs(centerPt.x - pt.x);
double yDist = Math.abs(centerPt.y - pt.y);
if (xDist > radius || yDist > radius)
continue;
if (xDist > innerBound || yDist > innerBound)
continue;
if (distSq(centerPt, pt) < radiusSq)
nearbyPtsSet.add(pt);
}
return nearbyPtsSet;
}

I've got my 2D/3D conversion working perfectly, how to do perspective

Although the context of this question is about making a 2d/3d game, the problem i have boils down to some math.
Although its a 2.5D world, lets pretend its just 2d for this question.
// xa: x-accent, the x coordinate of the projection
// mapP: a coordinate on a map which need to be projected
// _Dist_ values are constants for the projection, choosing them correctly will result in i.e. an isometric projection
xa = mapP.x * xDistX + mapP.y * xDistY;
ya = mapP.x * yDistX + mapP.y * yDistY;
xDistX and yDistX determine the angle of the x-axis, and xDistY and yDistY determine the angle of the y-axis on the projection (and also the size of the grid, but lets assume this is 1-pixel for simplicity).
x-axis-angle = atan(yDistX/xDistX)
y-axis-angle = atan(yDistY/yDistY)
a "normal" coordinate system like this
--------------- x
|
|
|
|
|
y
has values like this:
xDistX = 1;
yDistX = 0;
xDistY = 0;
YDistY = 1;
So every step in x direction will result on the projection to 1 pixel to the right end 0 pixels down. Every step in the y direction of the projection will result in 0 steps to the right and 1 pixel down.
When choosing the correct xDistX, yDistX, xDistY, yDistY, you can project any trimetric or dimetric system (which is why i chose this).
So far so good, when this is drawn everything turns out okay. If "my system" and mindset are clear, lets move on to perspective.
I wanted to add some perspective to this grid so i added some extra's like this:
camera = new MapPoint(60, 60);
dx = mapP.x - camera.x; // delta x
dy = mapP.y - camera.y; // delta y
dist = Math.sqrt(dx * dx + dy * dy); // dist is the distance to the camera, Pythagoras etc.. all objects must be in front of the camera
fac = 1 - dist / 100; // this formula determines the amount of perspective
xa = fac * (mapP.x * xDistX + mapP.y * xDistY) ;
ya = fac * (mapP.x * yDistX + mapP.y * yDistY );
Now the real hard part... what if you got a (xa,ya) point on the projection and want to calculate the original point (x,y).
For the first case (without perspective) i did find the inverse function, but how can this be done for the formula with the perspective. May math skills are not quite up to the challenge to solve this.
( I vaguely remember from a long time ago mathematica could create inverse function for some special cases... could it solve this problem? Could someone maybe try?)
The function you've defined doesn't have an inverse. Just as an example, as user207422 already pointed out anything that's 100 units away from the camera will get mapped to (xa,ya)=(0,0), so the inverse isn't uniquely defined.
More importantly, that's not how you calculate perspective. Generally the perspective scaling factor is defined to be viewdist/zdist where zdist is the perpendicular distance from the camera to the object and viewdist is a constant which is the distance from the camera to the hypothetical screen onto which everything is being projected. (See the diagram here, but feel free to ignore everything else on that page.) The scaling factor you're using in your example doesn't have the same behaviour.
Here's a stab at trying to convert your code into a correct perspective calculation (note I'm not simplifying to 2D; perspective is about projecting three dimensions to two, trying to simplify the problem to 2D is kind of pointless):
camera = new MapPoint(60, 60, 10);
camera_z = camera.x*zDistX + camera.y*zDistY + camera.z*zDistz;
// viewdist is the distance from the viewer's eye to the screen in
// "world units". You'll have to fiddle with this, probably.
viewdist = 10.0;
xa = mapP.x*xDistX + mapP.y*xDistY + mapP.z*xDistZ;
ya = mapP.x*yDistX + mapP.y*yDistY + mapP.z*yDistZ;
za = mapP.x*zDistX + mapP.y*zDistY + mapP.z*zDistZ;
zdist = camera_z - za;
scaling_factor = viewdist / zdist;
xa *= scaling_factor;
ya *= scaling_factor;
You're only going to return xa and ya from this function; za is just for the perspective calculation. I'm assuming the the "za-direction" points out of the screen, so if the pre-projection x-axis points towards the viewer then zDistX should be positive and vice-versa, and similarly for zDistY. For a trimetric projection you would probably have xDistZ==0, yDistZ<0, and zDistZ==0. This would make the pre-projection z-axis point straight up post-projection.
Now the bad news: this function doesn't have an inverse either. Any point (xa,ya) is the image of an infinite number of points (x,y,z). But! If you assume that z=0, then you can solve for x and y, which is possibly good enough.
To do that you'll have to do some linear algebra. Compute camera_x and camera_y similar to camera_z. That's the post-transformation coordinates of the camera. The point on the screen has post-tranformation coordinates (xa,ya,camera_z-viewdist). Draw a line through those two points, and calculate where in intersects the plane spanned by the vectors (xDistX, yDistX, zDistX) and (xDistY, yDistY, zDistY). In other words, you need to solve the equations:
x*xDistX + y*xDistY == s*camera_x + (1-s)*xa
x*yDistX + y*yDistY == s*camera_y + (1-s)*ya
x*zDistX + y*zDistY == s*camera_z + (1-s)*(camera_z - viewdist)
It's not pretty, but it will work.
I think that with your post i can solve the problem. Still, to clarify some questions:
Solving the problem in 2d is useless indeed, but this was only done to make the problem easier to grasp (for me and for the readers here). My program actually give's a perfect 3d projection (i checked it with 3d images rendered with blender). I did left something out about the inverse function though. The inverse function is only for coordinates between 0..camera.x * 0.5 and 0.. camera.y*0.5. So in my example between 0 and 30. But even then i have doubt's about my function.
In my projection the z-axis is always straight up, so to calculate the height of an object i only used the vieuwingangle. But since you cant actually fly or jumpt into the sky everything has only a 2d point. This also means that when you try to solve the x and y, the z really is 0.
I know not every funcion has an inverse, and some functions do, but only for a particular domain. My basic thought in this all was... if i can draw a grid using a function... every point on that grid maps to exactly one map-point. I can read the x and y coordinate so if i just had the correct function i would be able to calculate the inverse.
But there is no better replacement then some good solid math, and im very glad you took the time to give a very helpfull responce :).

Resources