How can I convert coordinates on a circle to coordinates on a square? - math

I'm currently working on a game in LBP2 that has modify the way a controller gives input. This question:
How can I convert coordinates on a square to coordinates on a circle?
Has helped me quite a lot with what I am doing, but I do have one problem. I need the inverse function of the one they give. They go from square -> circle, and I've tried searching all over for how to map a circle to a square.
The function given in the previous question is:
xCircle = xSquare * sqrt(1 - 0.5*ySquare^2)
yCircle = ySquare * sqrt(1 - 0.5*xSquare^2)
From Mapping a Square to a Circle
My question is given xCircle and yCircle... how do I find xSquare and ySquare?
I've tried all of the algebra I know, filled up two pages of notes, tried to get wolfram alpha to get the inverse functions, but this problem is beyond my abilities.
Thank you for taking a look.

x = ½ √( 2 + u² - v² + 2u√2 ) - ½ √( 2 + u² - v² - 2u√2 )
y = ½ √( 2 - u² + v² + 2v√2 ) - ½ √( 2 - u² + v² - 2v√2 )
Note on notation: I'm using x = xSquare , y = ySquare, u = xCircle and v = yCircle;
i.e. (u,v) are circular disc coordinates and (x,y) are square coordinates.
For a C++ implementation of the equations, go to
http://squircular.blogspot.com/2015/09/mapping-circle-to-square.html
See http://squircular.blogspot.com
for more example images.
Also, see http://arxiv.org/abs/1509.06344 for the proof/derivation
This mapping is the inverse of
u = x √( 1 - ½ y² )
v = y √( 1 - ½ x² )
P.S. The mapping is not unique. There are other mappings out there. The picture below illustrates the non-uniqueness of the mapping.

if you have xCircle and yCircle that means that you're on a circle with radius R = sqrt(xCircle^2 + yCircle^2). Now you need to extend that circle to a square with half-side = R,
if (xCircle < yCircle)
ySquare = R, xSquare = xCircle * R/yCircle
else
xSquare = R, ySquare = yCircle * R/xCircle
this is for the first quadrant, for others you need some trivial tweaking with the signs

There are many ways you could do this; here's one simple way.
Imagine a circle of radius R centred on the origin, and a square of side 2R centred on the origin, we want to map all of the points within and on the boundary of the circle (with coordinates (x,y)) to points within and on the boundary of the square. Note that we can also describe points within the circle using polar coordinates (r, ø) (that's supposed to be a phi), where
x = r cos ø,
y = r sin ø
(ie r^2 = x^2 + y^2 and r <= 1). Then imagine other coordinates x' = a(ø) x = a(ø) r cos ø, and y' = a(ø) y (ie, we decide that a won't depend on r).
In order to map the boundary of the circle (r = 1) to the boundary of the square (x' = R), we must have, for ø < 45deg, x' = a(ø) R cos ø = R, so we must have a(ø) = 1/cos ø. Similarly, for 45 < ø < 90 we must have the boundary of the circle map to y' = R, giving a(ø) = 1/sin ø in that region. Continuing round the circle, we see that a(ø) must always be positive, so the final mapping from the circle to the square is
x' = a(ø) x,
y' = a(ø) y
where
ø = |arctan y/x| = arctan |y/x|
and
a(ø) = 1/cos ø, when ø <= 45 deg (ie, when x < y), and
a(ø) = 1/sin ø, when ø > 45 deg.
That immediately gives you the mapping in the other direction. If you have coordinates (x', y') on the square (where x' <= R and y' <= R), then
x = x'/a(ø)
y = y'/a(ø)
with a(ø) as above.
A much simpler mapping, though, is to calculate the (r, ø) for the desired position on the circle, and map that to x' = r and y' = ø. That also maps every point in the circle into a rectangle, and vice versa, and might have better properties, depending on what you want to do.
So that's the real question: what is it you're actually aiming to do here?

I was implementing the solution above but the results are not satisfiying.
The square coordinates are not exact.
Here is a simple counter-example:
Consider the point (x,y)=(0.75, 1) on the square.
We map it to the circle with (u,v)=(0.53, 0.85) on the circle.
Applying the expression above we get the new square coordinates
(x',y')=(u/v,r)=(0.625543242, 1) with r=(u^2+v^2)^(1/2).
This point is close but not the expected precise solution.
I solved a root finding problem in order to get the inverse expression of the mapping from square to circle like above.
you need to solve the system equations like above:
I) u = x*(1-y^2/2)^(1/2)
II) v = y*(1-x^2/2)^(1/2)
One ends up with 8 root points as solution. One of the roots I implemented into Excel-VBA which I present here below and it works very fine.
' given the circle coordinates (u,v) caluclates the x coordinate on the square
Function circ2sqrX(u As Double, v As Double) As Double
Dim r As Double, signX As Double, u2 As Double, v2 As Double, uuvv As Double, temp1 As Double
u2 = u * u
v2 = v * v
r = Sqr(u2 + v2)
signX = 1
If v = 0 Or u = 0 Then
circ2sqrX = u
Exit Function
End If
If u < 0 Then
signX = -1
End If
If Abs(u) = Abs(v) And r = 1 Then
circ2sqrX = signX
Exit Function
End If
uuvv = (u2 - v2) * (u2 - v2) / 4
temp1 = 2 * Sqr(uuvv - u2 - v2 + 1)
circ2sqrX = -((temp1 - u2 + v2 - 2) * Sqr(temp1 + u2 - v2 + 2)) / (4 * u)
End Function
' given the circle coordinates (u,v) caluclates the y coordinate on the square
' make use of symetrie property
Function circ2sqrY(u As Double, v As Double) As Double
circ2sqrY=circ2sqrX(v,u)
End Function

Related

Positioning objects parallel with a mesh

I'm trying to align multiple line objects along a human body circumference depending on the orientation of the triangles from the mesh. I would like to put the lines parallel to the mesh. I correctly assign the position for the lines along the circumference, but I also need to add the rotation of the lines such that to be parallel with the body.
The body is a mesh formed by multiple triangles and every line is "linked" with a triangle.
All I have is:
3 points for the closest triangle from the mesh for every line
The normal of the triangle
The positions for the instantiated lines (2 points, start and end)
I need to calculate the angle for every X, Y, Z axes for the line such that the normal of the triangle is perpendicular with the line mesh. I don't know how to get the desired angle. I really appreciate if someone would like to help me.
input:
FVector TrianglePoints[3];
FVector Triangle_Normal; //Calculated as (B-A)^(C-A), where A,B,C are the points of the triangle
FVector linePosition; //I also have the start line and the endLine position if that helps
ouput:
//FRotator rotation(x,y,z), such that the triangle normal and the line object to be perpendicular.
An overview of the circumference line construction. Now the rotation is calculated using the Start position and End position for each line. When we cross some irregular parts of the mesh we want to rotate the lines correctly. Now the rotation is fixed, depending just on the line start and end position.
If I have understood correctly your goal, here is some related vector geometry:
A,B,C are the vertices of the triangle:
A = [xA, yA, zA],
B = [xB, yB, zB]
C = [xC, yC, zC]
K,L are the endpoints of the line-segment:
K = [xK, yK, zK]
L = [xL, yL, zL]
vectors are interpreted as row-vectors
by . I denote matrix multiplication
by x I denote cross product of 3D vectors
by t() I denote the transpose of a matrix
by | | I denote the norm (magnitude) of a vector
Goal: find the rotation matrix and rotation transformation of segment KL
around its midpoint, so that after rotation KL is parallel to the plane ABC
also, the rotation is the "minimal" angle rotation by witch we need to
rotate KL in order to make it parallel to ABC
AB = B - A
AC = C - A
KL = L - K
n = AB x AC
n = n / |n|
u = KL x n
u = u / |u|
v = n x u
cos = ( KL . t(v) ) / |KL|
sin = ( KL . t(n) ) / |KL|
U = [[ u[0], u[1], u[2] ],
[ v[0], v[1], v[2] ],
[ n[0], n[1], n[2] ],
R = [[1, 0, 0],
[0, cos, sin],
[0, -sin, cos]]
ROT = t(U).R.U
then, one can rotate the segment KL around its midpoint
M = (K + L)/2
Y = M + ROT (X - M)
Here is a python script version
A = np.array([0,0,0])
B = np.array([3,0,0])
C = np.array([2,3,0])
K = np.array([ -1,0,1])
L = np.array([ 2,2,2])
KL = L-K
U = np.empty((3,3), dtype=float)
U[2,:] = np.cross(B-A, C-A)
U[2,:] = U[2,:] / np.linalg.norm(U[2,:])
U[0,:] = np.cross(KL, U[2,:])
U[0,:] = U[0,:] / np.linalg.norm(U[0,:])
U[1,:] = np.cross(U[2,:], U[0,:])
norm_KL = np.linalg.norm(KL)
cos_ = KL.dot(U[1,:]) / norm_KL
sin_ = KL.dot(U[2,:]) / norm_KL
R = np.array([[1, 0, 0],
[0, cos_, sin_],
[0,-sin_, cos_]])
ROT = (U.T).dot(R.dot(U))
M = (K+L) / 2
K_rot = M + ROT.dot( K - M )
L_rot = M + ROT.dot( L - M )
print(L_rot)
print(K_rot)
print(L_rot-K_rot)
print((L_rot-K_rot).dot(U[2,:]))
A more inspired solution was to use a procedural mesh, generated at runtime, that have all the requirements that I need:
Continuously along multiple vertices
Easy to apply a UV map for texture tiling
Can be updated at runtime
Isn't hard to compute/work with it

Find line that is tangent to 2 given circles

I've got a situation in which I have 2 circles (C1 and C2)
and i need to find the line equation for the line that is tangent to both of these circles.
So as far as i'm aware, given a single point (P1) and C2's point and radius it is possible to quite easily get 2 possible points of tangency for C2 and P1 to make 2 line equations. But as i don't have P1, only the knowledge that the point will be one of a possible 2 points on C1, i'm not sure how to calculate this.
I assume it will be something along the lines of getting the 2 tangent line equations of C1 that are equal to the same of C2.
Both circles can have any radius, they could be the same or they could be hugely different. They will also never overlap (they can still touch though). And I'm looking for the 2 possible internal tangents.
Oh, and also, visuals would be very helpful haha :)
Let O be the intersection point between the line through the centers and the tangent.
Let d be the distance between the centers and h1, h2 be the distances between O and the centers. By similarity, these are proportional to the radii.
Hence,
h1 / h2 = r1 / r2 = m,
h1 + h2 = d,
giving
h1 = m d / (1 + m),
h2 = d / (1 + m).
Then the coordinates of O are found by interpolating between the centers
xo = (h2.x1 + h1.x2) / d
yo = (h2.y1 + h1.y2) / d
and the angle of the tangent is that of the line through the centers plus or minus the angle between this line and the tangent,
a = arctan((y2 - y1)/(x2 - x1)) +/- arcsin(r1 / h1).
You can write the implicit equation of the tangent as
cos(a).y - sin(a).x = cos(a).yo - sin(a).xo.
(source: imag.fr)
So we are going to use a homothetic transformation. If the circles C and C' have respectively centres O and O', and radius r and r', then we know there exists a unique homothetic transformation with centre J and ratio a, such that :
a = |JO|/|JO'| = r/r'
Noting AB is the vector from A to B, and |z| the norm of a vector z.
Hence you get J, knowing that it is between O and O' which we both already know.
Then with u the projection of JR on JO', and v the decomposition on its orthogonal, and considering the sine s and cosine c of the angle formed by O'JR, we have
|u| = |JR| * c
|v| = |JR| * s
c^2 + s^2 = 1
And finally because the triangle JRO' is right-angled in R :
s = r' / |JO|'
Putting all of this together, we get :
J = O + OO' / |OO'| * a / (a+1)
if |OJ| == r and |O'J| == r' then
return the orthogonal line to (OO') passing through J
|JR| = √( |JO'|^ - r'^2 )
s = r' / |JO'|
c = √( 1 - s^2 )
u = c * |JR| * OO' / |OO'|
w = (-u.y, u.x) % any orthogonal vector to u
v = s * |JR| * w / |w|
return lines corresponding to parametric equations J+t*(u+v) and J+t*(u-v)

Find the rotation angles of a triangle in 3D, given the coordinates of its vertices

I try to rotate and translate an equilateral triangle in 3D until his vertices reach some coordinates.
The vertices coordinates F,G,H and F',G',H' are known :
I was able to find the new centroid c' coordinates like this :
c'.x = ( F'.x + G'.x + H'.x ) / 3
c'.y = ( F'.y + G'.y + H'.y ) / 3
c'.z = ( F'.z + G'.z + H'.z ) / 3
So no problem to translate the triangle. But I can't find a way to calculate the rotations needed to put F'G'H' triangle in the right position...
I have to know by how much the triangle F'G'H' has to be rotated in degrees, around each axis (x,y,z), knowing that the rotations of the initial triangle are 0°.
By rotation for each axis, I'm talking about this:
Any ideas?
trick is to find the normal vectors of the triangles using cross product b4 and after rotations
v1 = (F.x - G.x, F.y - G.y, F.z - G.z)
v2 = (F.x - H.x, F.y - H.y, F.z - H.z)
n = cross_prod(v1, v2) # see http://en.wikipedia.org/wiki/Cross_product
n = n / norm(n) # normalize to unit vector
v'1 = (F'.x - G'.x, F'.y - G'.y, F'.z - G'.z)
v'2 = (F'.x - H'.x, F'.y - H'.y, F'.z - H'.z)
n' = cross_prod(v'1, v'2)
n' = n' / norm(n')
rot = arc_cos(n.x * n'.x + n.y * n'.y + n.z * n'.z)

Computing the 3D coordinates on a unit sphere from a 2D point

I have a square bitmap of a circle and I want to compute the normals of all the pixels in that circle as if it were a sphere of radius 1:
The sphere/circle is centered in the bitmap.
What is the equation for this?
Don't know much about how people program 3D stuff, so I'll just give the pure math and hope it's useful.
Sphere of radius 1, centered on origin, is the set of points satisfying:
x2 + y2 + z2 = 1
We want the 3D coordinates of a point on the sphere where x and y are known. So, just solve for z:
z = ±sqrt(1 - x2 - y2).
Now, let us consider a unit vector pointing outward from the sphere. It's a unit sphere, so we can just use the vector from the origin to (x, y, z), which is, of course, <x, y, z>.
Now we want the equation of a plane tangent to the sphere at (x, y, z), but this will be using its own x, y, and z variables, so instead I'll make it tangent to the sphere at (x0, y0, z0). This is simply:
x0x + y0y + z0z = 1
Hope this helps.
(OP):
you mean something like:
const int R = 31, SZ = power_of_two(R*2);
std::vector<vec4_t> p;
for(int y=0; y<SZ; y++) {
for(int x=0; x<SZ; x++) {
const float rx = (float)(x-R)/R, ry = (float)(y-R)/R;
if(rx*rx+ry*ry > 1) { // outside sphere
p.push_back(vec4_t(0,0,0,0));
} else {
vec3_t normal(rx,sqrt(1.-rx*rx-ry*ry),ry);
p.push_back(vec4_t(normal,1));
}
}
}
It does make a nice spherical shading-like shading if I treat the normals as colours and blit it; is it right?
(TZ)
Sorry, I'm not familiar with those aspects of C++. Haven't used the language very much, nor recently.
This formula is often used for "fake-envmapping" effect.
double x = 2.0 * pixel_x / bitmap_size - 1.0;
double y = 2.0 * pixel_y / bitmap_size - 1.0;
double r2 = x*x + y*y;
if (r2 < 1)
{
// Inside the circle
double z = sqrt(1 - r2);
.. here the normal is (x, y, z) ...
}
Obviously you're limited to assuming all the points are on one half of the sphere or similar, because of the missing dimension. Past that, it's pretty simple.
The middle of the circle has a normal facing precisely in or out, perpendicular to the plane the circle is drawn on.
Each point on the edge of the circle is facing away from the middle, and thus you can calculate the normal for that.
For any point between the middle and the edge, you use the distance from the middle, and some simple trig (which eludes me at the moment). A lerp is roughly accurate at some points, but not quite what you need, since it's a curve. Simple curve though, and you know the beginning and end values, so figuring them out should only take a simple equation.
I think I get what you're trying to do: generate a grid of depth data for an image. Sort of like ray-tracing a sphere.
In that case, you want a Ray-Sphere Intersection test:
http://www.siggraph.org/education/materials/HyperGraph/raytrace/rtinter1.htm
Your rays will be simple perpendicular rays, based off your U/V coordinates (times two, since your sphere has a diameter of 2). This will give you the front-facing points on the sphere.
From there, calculate normals as below (point - origin, the radius is already 1 unit).
Ripped off from the link above:
You have to combine two equations:
Ray: R(t) = R0 + t * Rd , t > 0 with R0 = [X0, Y0, Z0] and Rd = [Xd, Yd, Zd]
Sphere: S = the set of points[xs, ys, zs], where (xs - xc)2 + (ys - yc)2 + (zs - zc)2 = Sr2
To do this, calculate your ray (x * pixel / width, y * pixel / width, z: 1), then:
A = Xd^2 + Yd^2 + Zd^2
B = 2 * (Xd * (X0 - Xc) + Yd * (Y0 - Yc) + Zd * (Z0 - Zc))
C = (X0 - Xc)^2 + (Y0 - Yc)^2 + (Z0 - Zc)^2 - Sr^2
Plug into quadratic equation:
t0, t1 = (- B + (B^2 - 4*C)^1/2) / 2
Check discriminant (B^2 - 4*C), and if real root, the intersection is:
Ri = [xi, yi, zi] = [x0 + xd * ti , y0 + yd * ti, z0 + zd * ti]
And the surface normal is:
SN = [(xi - xc)/Sr, (yi - yc)/Sr, (zi - zc)/Sr]
Boiling it all down:
So, since we're talking unit values, and rays that point straight at Z (no x or y component), we can boil down these equations greatly:
Ray:
X0 = 2 * pixelX / width
Y0 = 2 * pixelY / height
Z0 = 0
Xd = 0
Yd = 0
Zd = 1
Sphere:
Xc = 1
Yc = 1
Zc = 1
Factors:
A = 1 (unit ray)
B
= 2 * (0 + 0 + (0 - 1))
= -2 (no x/y component)
C
= (X0 - 1) ^ 2 + (Y0 - 1) ^ 2 + (0 - 1) ^ 2 - 1
= (X0 - 1) ^ 2 + (Y0 - 1) ^ 2
Discriminant
= (-2) ^ 2 - 4 * 1 * C
= 4 - 4 * C
From here:
If discriminant < 0:
Z = ?, Normal = ?
Else:
t = (2 + (discriminant) ^ 1 / 2) / 2
If t < 0 (hopefully never or always the case)
t = -t
Then:
Z: t
Nx: Xi - 1
Ny: Yi - 1
Nz: t - 1
Boiled farther still:
Intuitively it looks like C (X^2 + Y^2) and the square-root are the most prominent figures here. If I had a better recollection of my math (in particular, transformations on exponents of sums), then I'd bet I could derive this down to what Tom Zych gave you. Since I can't, I'll just leave it as above.

correcting fisheye distortion programmatically

BOUNTY STATUS UPDATE:
I discovered how to map a linear lens, from destination coordinates to source coordinates.
How do you calculate the radial distance from the centre to go from fisheye to rectilinear?
1). I actually struggle to reverse it, and to map source coordinates to destination coordinates. What is the inverse, in code in the style of the converting functions I posted?
2). I also see that my undistortion is imperfect on some lenses - presumably those that are not strictly linear. What is the equivalent to-and-from source-and-destination coordinates for those lenses? Again, more code than just mathematical formulae please...
Question as originally stated:
I have some points that describe positions in a picture taken with a fisheye lens.
I want to convert these points to rectilinear coordinates. I want to undistort the image.
I've found this description of how to generate a fisheye effect, but not how to reverse it.
There's also a blog post that describes how to use tools to do it; these pictures are from that:
(1) : SOURCE Original photo link
Input : Original image with fish-eye distortion to fix.
(2) : DESTINATION Original photo link
Output : Corrected image (technically also with perspective correction, but that's a separate step).
How do you calculate the radial distance from the centre to go from fisheye to rectilinear?
My function stub looks like this:
Point correct_fisheye(const Point& p,const Size& img) {
// to polar
const Point centre = {img.width/2,img.height/2};
const Point rel = {p.x-centre.x,p.y-centre.y};
const double theta = atan2(rel.y,rel.x);
double R = sqrt((rel.x*rel.x)+(rel.y*rel.y));
// fisheye undistortion in here please
//... change R ...
// back to rectangular
const Point ret = Point(centre.x+R*cos(theta),centre.y+R*sin(theta));
fprintf(stderr,"(%d,%d) in (%d,%d) = %f,%f = (%d,%d)\n",p.x,p.y,img.width,img.height,theta,R,ret.x,ret.y);
return ret;
}
Alternatively, I could somehow convert the image from fisheye to rectilinear before finding the points, but I'm completely befuddled by the OpenCV documentation. Is there a straightforward way to do it in OpenCV, and does it perform well enough to do it to a live video feed?
The description you mention states that the projection by a pin-hole camera (one that does not introduce lens distortion) is modeled by
R_u = f*tan(theta)
and the projection by common fisheye lens cameras (that is, distorted) is modeled by
R_d = 2*f*sin(theta/2)
You already know R_d and theta and if you knew the camera's focal length (represented by f) then correcting the image would amount to computing R_u in terms of R_d and theta. In other words,
R_u = f*tan(2*asin(R_d/(2*f)))
is the formula you're looking for. Estimating the focal length f can be solved by calibrating the camera or other means such as letting the user provide feedback on how well the image is corrected or using knowledge from the original scene.
In order to solve the same problem using OpenCV, you would have to obtain the camera's intrinsic parameters and lens distortion coefficients. See, for example, Chapter 11 of Learning OpenCV (don't forget to check the correction). Then you can use a program such as this one (written with the Python bindings for OpenCV) in order to reverse lens distortion:
#!/usr/bin/python
# ./undistort 0_0000.jpg 1367.451167 1367.451167 0 0 -0.246065 0.193617 -0.002004 -0.002056
import sys
import cv
def main(argv):
if len(argv) < 10:
print 'Usage: %s input-file fx fy cx cy k1 k2 p1 p2 output-file' % argv[0]
sys.exit(-1)
src = argv[1]
fx, fy, cx, cy, k1, k2, p1, p2, output = argv[2:]
intrinsics = cv.CreateMat(3, 3, cv.CV_64FC1)
cv.Zero(intrinsics)
intrinsics[0, 0] = float(fx)
intrinsics[1, 1] = float(fy)
intrinsics[2, 2] = 1.0
intrinsics[0, 2] = float(cx)
intrinsics[1, 2] = float(cy)
dist_coeffs = cv.CreateMat(1, 4, cv.CV_64FC1)
cv.Zero(dist_coeffs)
dist_coeffs[0, 0] = float(k1)
dist_coeffs[0, 1] = float(k2)
dist_coeffs[0, 2] = float(p1)
dist_coeffs[0, 3] = float(p2)
src = cv.LoadImage(src)
dst = cv.CreateImage(cv.GetSize(src), src.depth, src.nChannels)
mapx = cv.CreateImage(cv.GetSize(src), cv.IPL_DEPTH_32F, 1)
mapy = cv.CreateImage(cv.GetSize(src), cv.IPL_DEPTH_32F, 1)
cv.InitUndistortMap(intrinsics, dist_coeffs, mapx, mapy)
cv.Remap(src, dst, mapx, mapy, cv.CV_INTER_LINEAR + cv.CV_WARP_FILL_OUTLIERS, cv.ScalarAll(0))
# cv.Undistort2(src, dst, intrinsics, dist_coeffs)
cv.SaveImage(output, dst)
if __name__ == '__main__':
main(sys.argv)
Also note that OpenCV uses a very different lens distortion model to the one in the web page you linked to.
(Original poster, providing an alternative)
The following function maps destination (rectilinear) coordinates to source (fisheye-distorted) coordinates. (I'd appreciate help in reversing it)
I got to this point through trial-and-error: I don't fundamentally grasp why this code is working, explanations and improved accuracy appreciated!
def dist(x,y):
return sqrt(x*x+y*y)
def correct_fisheye(src_size,dest_size,dx,dy,factor):
""" returns a tuple of source coordinates (sx,sy)
(note: values can be out of range)"""
# convert dx,dy to relative coordinates
rx, ry = dx-(dest_size[0]/2), dy-(dest_size[1]/2)
# calc theta
r = dist(rx,ry)/(dist(src_size[0],src_size[1])/factor)
if 0==r:
theta = 1.0
else:
theta = atan(r)/r
# back to absolute coordinates
sx, sy = (src_size[0]/2)+theta*rx, (src_size[1]/2)+theta*ry
# done
return (int(round(sx)),int(round(sy)))
When used with a factor of 3.0, it successfully undistorts the images used as examples (I made no attempt at quality interpolation):
Dead link
(And this is from the blog post, for comparison:)
If you think your formulas are exact, you can comput an exact formula with trig, like so:
Rin = 2 f sin(w/2) -> sin(w/2)= Rin/2f
Rout= f tan(w) -> tan(w)= Rout/f
(Rin/2f)^2 = [sin(w/2)]^2 = (1 - cos(w))/2 -> cos(w) = 1 - 2(Rin/2f)^2
(Rout/f)^2 = [tan(w)]^2 = 1/[cos(w)]^2 - 1
-> (Rout/f)^2 = 1/(1-2[Rin/2f]^2)^2 - 1
However, as #jmbr says, the actual camera distortion will depend on the lens and the zoom. Rather than rely on a fixed formula, you might want to try a polynomial expansion:
Rout = Rin*(1 + A*Rin^2 + B*Rin^4 + ...)
By tweaking first A, then higher-order coefficients, you can compute any reasonable local function (the form of the expansion takes advantage of the symmetry of the problem). In particular, it should be possible to compute initial coefficients to approximate the theoretical function above.
Also, for good results, you will need to use an interpolation filter to generate your corrected image. As long as the distortion is not too great, you can use the kind of filter you would use to rescale the image linearly without much problem.
Edit: as per your request, the equivalent scaling factor for the above formula:
(Rout/f)^2 = 1/(1-2[Rin/2f]^2)^2 - 1
-> Rout/f = [Rin/f] * sqrt(1-[Rin/f]^2/4)/(1-[Rin/f]^2/2)
If you plot the above formula alongside tan(Rin/f), you can see that they are very similar in shape. Basically, distortion from the tangent becomes severe before sin(w) becomes much different from w.
The inverse formula should be something like:
Rin/f = [Rout/f] / sqrt( sqrt(([Rout/f]^2+1) * (sqrt([Rout/f]^2+1) + 1) / 2 )
I blindly implemented the formulas from here, so I cannot guarantee it would do what you need.
Use auto_zoom to get the value for the zoom parameter.
def dist(x,y):
return sqrt(x*x+y*y)
def fisheye_to_rectilinear(src_size,dest_size,sx,sy,crop_factor,zoom):
""" returns a tuple of dest coordinates (dx,dy)
(note: values can be out of range)
crop_factor is ratio of sphere diameter to diagonal of the source image"""
# convert sx,sy to relative coordinates
rx, ry = sx-(src_size[0]/2), sy-(src_size[1]/2)
r = dist(rx,ry)
# focal distance = radius of the sphere
pi = 3.1415926535
f = dist(src_size[0],src_size[1])*factor/pi
# calc theta 1) linear mapping (older Nikon)
theta = r / f
# calc theta 2) nonlinear mapping
# theta = asin ( r / ( 2 * f ) ) * 2
# calc new radius
nr = tan(theta) * zoom
# back to absolute coordinates
dx, dy = (dest_size[0]/2)+rx/r*nr, (dest_size[1]/2)+ry/r*nr
# done
return (int(round(dx)),int(round(dy)))
def fisheye_auto_zoom(src_size,dest_size,crop_factor):
""" calculate zoom such that left edge of source image matches left edge of dest image """
# Try to see what happens with zoom=1
dx, dy = fisheye_to_rectilinear(src_size, dest_size, 0, src_size[1]/2, crop_factor, 1)
# Calculate zoom so the result is what we wanted
obtained_r = dest_size[0]/2 - dx
required_r = dest_size[0]/2
zoom = required_r / obtained_r
return zoom
I took what JMBR did and basically reversed it. He took the radius of the distorted image (Rd, that is, the distance in pixels from the center of the image) and found a formula for Ru, the radius of the undistorted image.
You want to go the other way. For each pixel in the undistorted (processed image), you want to know what the corresponding pixel is in the distorted image.
In other words, given (xu, yu) --> (xd, yd). You then replace each pixel in the undistorted image with its corresponding pixel from the distorted image.
Starting where JMBR did, I do the reverse, finding Rd as a function of Ru. I get:
Rd = f * sqrt(2) * sqrt( 1 - 1/sqrt(r^2 +1))
where f is the focal length in pixels (I'll explain later), and r = Ru/f.
The focal length for my camera was 2.5 mm. The size of each pixel on my CCD was 6 um square. f was therefore 2500/6 = 417 pixels. This can be found by trial and error.
Finding Rd allows you to find the corresponding pixel in the distorted image using polar coordinates.
The angle of each pixel from the center point is the same:
theta = arctan( (yu-yc)/(xu-xc) ) where xc, yc are the center points.
Then,
xd = Rd * cos(theta) + xc
yd = Rd * sin(theta) + yc
Make sure you know which quadrant you are in.
Here is the C# code I used
public class Analyzer
{
private ArrayList mFisheyeCorrect;
private int mFELimit = 1500;
private double mScaleFESize = 0.9;
public Analyzer()
{
//A lookup table so we don't have to calculate Rdistorted over and over
//The values will be multiplied by focal length in pixels to
//get the Rdistorted
mFisheyeCorrect = new ArrayList(mFELimit);
//i corresponds to Rundist/focalLengthInPixels * 1000 (to get integers)
for (int i = 0; i < mFELimit; i++)
{
double result = Math.Sqrt(1 - 1 / Math.Sqrt(1.0 + (double)i * i / 1000000.0)) * 1.4142136;
mFisheyeCorrect.Add(result);
}
}
public Bitmap RemoveFisheye(ref Bitmap aImage, double aFocalLinPixels)
{
Bitmap correctedImage = new Bitmap(aImage.Width, aImage.Height);
//The center points of the image
double xc = aImage.Width / 2.0;
double yc = aImage.Height / 2.0;
Boolean xpos, ypos;
//Move through the pixels in the corrected image;
//set to corresponding pixels in distorted image
for (int i = 0; i < correctedImage.Width; i++)
{
for (int j = 0; j < correctedImage.Height; j++)
{
//which quadrant are we in?
xpos = i > xc;
ypos = j > yc;
//Find the distance from the center
double xdif = i-xc;
double ydif = j-yc;
//The distance squared
double Rusquare = xdif * xdif + ydif * ydif;
//the angle from the center
double theta = Math.Atan2(ydif, xdif);
//find index for lookup table
int index = (int)(Math.Sqrt(Rusquare) / aFocalLinPixels * 1000);
if (index >= mFELimit) index = mFELimit - 1;
//calculated Rdistorted
double Rd = aFocalLinPixels * (double)mFisheyeCorrect[index]
/mScaleFESize;
//calculate x and y distances
double xdelta = Math.Abs(Rd*Math.Cos(theta));
double ydelta = Math.Abs(Rd * Math.Sin(theta));
//convert to pixel coordinates
int xd = (int)(xc + (xpos ? xdelta : -xdelta));
int yd = (int)(yc + (ypos ? ydelta : -ydelta));
xd = Math.Max(0, Math.Min(xd, aImage.Width-1));
yd = Math.Max(0, Math.Min(yd, aImage.Height-1));
//set the corrected pixel value from the distorted image
correctedImage.SetPixel(i, j, aImage.GetPixel(xd, yd));
}
}
return correctedImage;
}
}
I found this pdf file and I have proved that the maths are correct (except for the line vd = *xd**fv+v0 which should say vd = **yd**+fv+v0).
http://perception.inrialpes.fr/CAVA_Dataset/Site/files/Calibration_OpenCV.pdf
It does not use all of the latest co-efficients that OpenCV has available but I am sure that it could be adapted fairly easily.
double k1 = cameraIntrinsic.distortion[0];
double k2 = cameraIntrinsic.distortion[1];
double p1 = cameraIntrinsic.distortion[2];
double p2 = cameraIntrinsic.distortion[3];
double k3 = cameraIntrinsic.distortion[4];
double fu = cameraIntrinsic.focalLength[0];
double fv = cameraIntrinsic.focalLength[1];
double u0 = cameraIntrinsic.principalPoint[0];
double v0 = cameraIntrinsic.principalPoint[1];
double u, v;
u = thisPoint->x; // the undistorted point
v = thisPoint->y;
double x = ( u - u0 )/fu;
double y = ( v - v0 )/fv;
double r2 = (x*x) + (y*y);
double r4 = r2*r2;
double cDist = 1 + (k1*r2) + (k2*r4);
double xr = x*cDist;
double yr = y*cDist;
double a1 = 2*x*y;
double a2 = r2 + (2*(x*x));
double a3 = r2 + (2*(y*y));
double dx = (a1*p1) + (a2*p2);
double dy = (a3*p1) + (a1*p2);
double xd = xr + dx;
double yd = yr + dy;
double ud = (xd*fu) + u0;
double vd = (yd*fv) + v0;
thisPoint->x = ud; // the distorted point
thisPoint->y = vd;
This can be solved as an optimization problem. Simply draw on curves in images that are supposed to be straight lines. Store the contour points for each of those curves. Now we can solve the fish eye matrix as a minimization problem. Minimize the curve in points and that will give us a fisheye matrix. It works.
It can be done manually by adjusting the fish eye matrix using trackbars! Here is a fish eye GUI code using OpenCV for manual calibration.

Resources