I am trying to curve fit 5 points in C. I have used this code from a previous post (Can sombody simplify this equation for me?) to do 4 points, but now I need to add another point.
// Input data: arrays x[] and y[]
// x[1],x[2],x[3],x[4] - X values
// y[1],y[2],y[3],y[4] - Y values
// Calculations
A = 0
B = 0
C = 0
D = 0
S1 = x[1] + x[2] + x[3] + x[4]
S2 = x[1]*x[2] + x[1]*x[3] + x[1]*x[4] + x[2]*x[3] + x[2]*x[4] + x[3]*x[4]
S3 = x[1]*x[2]*x[3] + x[1]*x[2]*x[4] + x[1]*x[3]*x[4] + x[2]*x[3]*x[4]
for i = 1 to 4 loop
C0 = y[i]/(((4*x[i]-3*S1)*x[i]+2*S2)*x[i]-S3)
C1 = C0*(S1 - x[i])
C2 = S2*C0 - C1*x[i]
C3 = S3*C0 - C2*x[i]
A = A + C0
B = B - C1
C = C + C2
D = D - C3
end-loop
// Result: A, B, C, D
I have been trying to covert this to a 5 point curve fit, but am having trouble figuring out what goes inside the loop:
// Input data: arrays x[] and y[]
// x[1],x[2],x[3],x[4],x[5] - X values
// y[1],y[2],y[3],y[4],y[5] - Y values
// Calculations
A = 0
B = 0
C = 0
D = 0
E = 0
S1 = x[1] + x[2] + x[3] + x[4]
S2 = x[1]*x[2] + x[1]*x[3] + x[1]*x[4] + x[2]*x[3] + x[2]*x[4] + x[3]*x[4]
S3 = x[1]*x[2]*x[3] + x[1]*x[2]*x[4] + x[1]*x[3]*x[4] + x[2]*x[3]*x[4]
S4 = x[1]*x[2]*x[3]*x[4] + x[1]*x[2]*x[3]*[5] + x[1]*x[2]*x[4]*[5] + x[1]*x[3]*x[4]*[5] + x[2]*x[3]*x[4]*[5]
for i = 1 to 4 loop
C0 = ??
C1 = ??
C2 = ??
C3 = ??
C4 = ??
A = A + C0
B = B - C1
C = C + C2
D = D - C3
E = E + C4
end-loop
// Result: A, B, C, D, E
any help in filling out the C0...C4 would be appreciated. I know this has to do with the matrices but I have not been able to figure it out. examples with pseudo code or real code would be most helpful.
thanks
I refuse to miss this opportunity to generalize. :)
Instead, we're going to learn a little bit about Lagrange polynomials and the Newton Divided Difference Method of their computation.
Lagrange Polynomials
Given n+1 data points, the interpolating polynomial is
where l_j(i) is
.
What this means is that we can find the polynomial approximating the n+1 points, regardless of spacing, etc, by just summing these polynomials. However, this is a bit of a pain and I wouldn't want to do it in C. Let's take a look at Newton Polynomials.
Newton Polynomials
Same start, given n+1 data points, the approximating polynomial is going to be
where each n(x) is
with a coefficient of
, being the divided difference.
The final form end's up looking like
.
As you can see, the formula is pretty easy given the divided difference values. You just do each new divided difference and multiply by each point so far. It should be noted that you'll end up with a polynomial of degree n from n+1 points.
Divided Difference
All that's left is to define the divided difference which is really best explained by these two pictures:
and
.
With this information, a C implementation should be reasonable to do. I hope this helps and I hope you learned something! :)
If the x values are equally spaced with x2-x1=h, x3-x2=h, x4-x3=h and x5-x4=h then
C0 = y1;
C1 = -(25*y1-48*y2+36*y3-16*y4+3*y5)/(12*h);
C2 = (35*y1-104*y2+114*y3-56*y4+11*y5)/(24*h*h);
C3 = -(5*y1-18*y2+24*y3-14*y4+3*y5)/(12*h*h*h);
C4 = (y1-4*y2+6*y3-4*y4+y5)/(24*h*h*h*h);
y(x) = C0+C1*(x-x1)+C2*(x-x1)^2+C3*(x-x1)^3+C4*(x-x1)^4
// where `^` denotes exponentiation (and not XOR).
Related
I would like to calculate the bisection of two 3D lines which have an intersecting point. The lines are sympy lines defined by a point and a direction vector. How can I find the equation of the two lines which are the bisection of them?
Let lines are defined as A + t * dA, B + s * dB where A, B are base points and dA, dB are normalized direction vectors.
If it is guaranteed that lines have intersection, it could be found using dot product approach (adapted from skew line minimal distance algorithm):
u = A - B
b = dot(dA, dB)
if abs(b) == 1: # better check with some tolerance
lines are parallel
d = dot(dA, u)
e = dot(dB, u)
t_intersect = (b * e - d) / (1 - b * b)
P = A + t_intersect * dA
Now about bisectors:
bis1 = P + v * normalized(dA + dB)
bis2 = P + v * normalized(dA - dB)
Quick check for 2D case
k = Sqrt(1/5)
A = (3,1) dA = (-k,2k)
B = (1,1) dB = (k,2k)
u = (2,0)
b = -k^2+4k2 = 3k^2=3/5
d = -2k e = 2k
t = (b * e - d) / (1 - b * b) =
(6/5*k+2*k) / (16/25) = 16/5*k * 25/16 = 5*k
Px = 3 - 5*k^2 = 2
Py = 1 + 10k^2 = 3
normalized(dA+dB=(0,4k)) = (0,1)
normalized(dA-dB=(-2k,0)) = (-1,0)
Python implementation:
from sympy.geometry import Line3D, Point3D, intersection
# Normalize direction vectors:
def normalize(vector: list):
length = (vector[0]**2 + vector[1]**2 + vector[2]**2)**0.5
vector = [i/length for i in vector]
return vector
# Example points for creating two lines which intersect at A
A = Point3D(1, 1, 1)
B = Point3D(0, 2, 1)
l1 = Line3D(A, direction_ratio=[1, 0, 0])
l2 = Line3D(A, B)
d1 = normalize(l1.direction_ratio)
d2 = normalize(l2.direction_ratio)
p = intersection(l1, l2)[0] # Point3D of intersection between the two lines
bis1 = Line3D(p, direction_ratio=[d1[i]+d2[i] for i in range(3)])
bis2 = Line3D(p, direction_ratio=[d1[i]-d2[i] for i in range(3)])
I have this triangle:
I'm trying to get the vertex value highlighted in the green circle in order to draw that red line. Is there any equation that I can use to extract that value?
The centroid vertex G = (x=5.5, y=1.5)
The other vertex B = (x=0, y=1)
and the last vertex C = (x=7, y=0)
Any help would be appreciated. I know it might be a 5th grade math but I can't think of a way to calculate this point.
If you throw away the majority of the triangle and just keep the vector B->G and the vector B->C then this problem shows itself to be a "vector projection" problem.
These are solved analytically using the dot product of the 2 vectors and are well documented elsewhere.
Took me 2 days to figure this out, you basically need to get the slopes for the base vector and the altitude vector (centroid), then solve this equation: y = m * x + b for both vectors (the base + altitude). Then you'll get 2 different equations that you need to use substitution to get the x first then apply that value to the 2nd equation to get the y. For more information watch this youtube tutorial:
https://www.youtube.com/watch?v=VuEbWkF5lcM
Here's the solution in PHP (pseudo) if anyone is interested:
//slope of base
$m1 = getSlope(baseVector);
//slope of altitude (invert and divide it by 1)
$m2 = 1/-$m1;
//points
$x1 = $baseVector->x;
$y1 = $baseVector->y;
//Centroid vertex
$x2 = $center['x'];
$y2 = $center['y'];
//altitude equation: y = m * x + b
//eq1: y1 = (m1 * x1) + b1 then find b1
$b1 = -($m1 * $x1) + $y1;
//equation: y = ($m1 * x) + $b1
//eq2: y2 = (m2 * x2) + b2 then find b2
$b2 = -($m2 * $x2) + $y2;
//equation: y = ($m2 * x) + $b2;
//substitute eq1 into eq2 and find x
//merge the equations (move the Xs to the left side and numbers on the right side)
$Xs = $m1 - $m2; //left side (number of Xs)
$Bs = $b2 - $b1; //right side
$x = $Bs / $Xs; //get x number
$y = ($m2 * $x) + $b2; //get y number
i have a setup like this:
2 coordinate systems. (x,y) is the main coordinate system and (x',y') is a coordinate system that lives inside (x,y). The system (x',y') is defined by the points x1 or x2 and if i move these 2 points around then (x',y') moves accordingly. The origin of (x',y') is defined as the middle of the vector going from x1 to x2, and the y' axis is the normal vector on x1->x2 going through the origin. If i have a point x3 defined in (x',y') and i move either of x1 or x2 to make the origin shift place, how do i then move x3 accordingly such that it maintains its position in the new (x',y') ?
And how do i make a transformation which always converts a point in (x,y) to a point in (x',y') nomatter how x1 and x2 have been set?
I was thinking that if i had more points than just the one i am moving (x1 or x2) i guess i could try to estimate theta, tx, ty of the transformation
[x2'] [cos(theta) , sin(theta), tx][x2]
[y2'] = [-sin(theta), cos(theta), ty][y2]
[ 1 ] [ 0 , 0 , 1 ][1 ]
and just apply that estimated transformation to x3 and i would be good...mmm but i think i would need 3 points in order to estimate theta, tx and ty right?
I mean i could estimate using some least squares approach...but 3 unknowns requires 3 coordinate sets right?
I tried to implement this and calculate an example. I hope you understand the syntax. Its not really giving me what i expect:
import math
import numpy as np
x1=[ 0,10]
x2=[10,20]
rx = x2[0] - x1[0]
ry = x2[1] - x1[1]
rlen = math.sqrt(rx*rx+ry*ry)
c = rx / rlen
s = ry / rlen
dx = - ( x1[0] + x2[0] )/2 # changing the sign to be negative seems to
dy = - ( x1[1] + x2[1] )/2 # rectify translation. Rotation still is wrong
M = np.array([[c, -s, 0],[s, c, 0],[dx, dy, 1]])
print( np.dot(x2 + [1],M) )
# Yields -> [ 15.92031022 -8.63603897 1. ] and should yield [5,0,1]
Since I am trying to transform the x2 coordinate, should the result then not have the value 0 in the y-component since its located in the x-axis?
Ok, I tried doing the implementation for x3 from dynamic1 to dynamic2 which the check is that x3 should end up with the same coordinate in both d1 and d2. I did that as you suggested, but I do not get the same coordinate in both d1 and d2. Did i misunderstand something?
import math
import numpy as np
x1=[ 1,1]
x2=[ 7,9]
x3=[4,3]
rx = (x2[0] - x1[0])
ry = (x2[1] - x1[1])
rlen = math.sqrt( rx*rx + ry*ry )
c = rx / rlen
s = ry / rlen
dx = ( x1[0] + x2[0] )/2
dy = ( x1[1] + x2[1] )/2
M = np.array([[c, -s, 0],[s, c, 0],[-dx*c-dy*s, dx*s-dy*c, 1]])
Minv = np.array([[c, s, 0],[-s, c, 0],[dx, dy, 1]])
x1new=[ 1,1]
x2new=[ 17,4]
rxnew = (x2new[0] - x1new[0])
rynew = (x2new[1] - x1new[1])
rlennew = math.sqrt( rxnew*rxnew + rynew*rynew )
cnew = rxnew / rlennew
snew = rynew / rlennew
dxnew = ( x1new[0] + x2new[0] )/2
dynew = ( x1new[1] + x2new[1] )/2
Mnew = np.array([[cnew, -snew, 0],[snew, cnew, 0],[-dxnew*cnew-dynew*snew, dxnew*snew-dynew*cnew, 1]])
Mnewinv = np.array([[cnew, snew, 0],[-snew, cnew, 0],[dxnew, dynew, 1]])
M_dyn1_to_dyn2 = np.dot(Minv,Mnew)
print( np.dot(x3 + [1], M) )
print( np.dot(x3 + [1], M_dyn1_to_dyn2))
#yields these 2 outputs which should be the same:
[-1.6 -1.2 1. ]
[-3.53219692 8.29298408 1. ]
Edit. Matrix correction.
To translate coordinates from static system to (x1,x2) defined one, you have to apply affine transformation.
Matrix of this transformation M consists of shift matrix S and rotation about origin R.
Matrix M is combination of S and R:
c -s 0
M = s c 0
-dx*c-dy*s dx*s-dy*c 1
Here c and s are cosine and sine of rotation angle, their values are respectively x- and y- components of unit (normalized) vector x1x2.
rx = x2.x - x1.x
ry = x2.y - x1.y
len = Sqrt(rx*rx+ry*ry)
c = rx / Len
s = ry / Len
And shift components:
dx = (x1.x + x2.x)/2
dy = (x1.y + x2.y)/2
To translate (xx,yy) coordinates from static system to rotate one, we have to find
xx' = xx*c+yy*s-dx*c-dy*s = c*(xx-dx) + s*(yy-dy)
yy' = -xx*s+yy*c+dx*s-dy*c = -s*(xx-dx) + c*(yy-dy)
Quick check:
X1 = (1,1)
X2 = (7,9)
dx = 4
dy = 5
rx = 6
ry = 8
Len = 10
c = 0.6
s = 0.8
for point (4,5):
xx-dx = 0
yy-dy = 0
xx',yy' = (0, 0) - right
for point X2 =(7,9):
xx-dx = 3
yy-dy = 4
xx' = 0.6*3 + 0.8*4 = 5 -right
yy' = -0.8*3 + 0.6*4 = 0 -right
P.S. Note that matrix to transform dyn.coordinates to static ones is inverse of M and it is simpler:
c s 0
M' = -s c 0
dx dy 1
P.P.S. You need three pairs of corresponding points to define general affine transformations. It seems here you don't need scaling and sheer, so you may determine needed transform with your x1,x2 points
I think you need double dimension array to save and set your value in that
the structure gonna be like this
=============|========|========|
index number |x |y |
=============|========|========|
first point | [0][0] | [0][1] |
second point | [1][0] | [1][1] |
third point | [2][0] | [2][1] |
=============|========|========|
I will use java in my answer
//declare the double dimension array
double matrix[][] = new double[3][2];
//setting location first point, x
matrix[0][0] = 1;
//setting location first point, y
matrix[0][1] = 1;
//fill with your formula, i only give example
//fill second point with first point and plus 1
//setting location second point, x
matrix[1][0] = matrix[0][0] + 1;
//setting location second point, y
matrix[1][1] = matrix[0][1] + 1;
//fill with your formula, i only give example
//fill third point with second point and plus 1
//setting location third point, x
matrix[2][0] = matrix[1][0] + 1;
//setting location third point, y
matrix[2][1] = matrix[1][1] + 1;
how can i calculate the polynomial that has the tangent lines (1) y = x where x = 1, and (2) y = 1 where x = 365
I realize this may not be the proper forum but I figured somebody here could answer this in jiffy.
Also, I am not looking for an algorithm to answer this. I'd just like like to see the process.
Thanks.
I guess I should have mentioned that i'm writing an algorithm for scaling the y-axis of flotr graph
The specification of the curve can be expressed as four constraints:
y(1) = 1, y'(1) = 1 => tangent is (y=x) when x=1
y(365) = 1, y'(365) = 0 => tangent is (y=1) when x=365
We therefore need a family of curves with at least four degrees of freedom to match these constraints; the simplest type of polynomial is a cubic,
y = a*x^3 + b*x^2 + c*x + d
y' = 3*a*x^2 + 2*b*x + c
and the constraints give the following equations for the parameters:
a + b + c + d = 1
3*a + 2*b + c = 1
48627125*a + 133225*b + 365*c + d = 1
399675*a + 730*b + c = 0
I'm too old and too lazy to solve these myself, so I googled a linear equation solver to give the answer:
a = 1/132496, b = -731/132496, c = 133955/132496, d = -729/132496
I will post this type of question in mathoverflow.net next time. thanks
my solution in javascript was to adapt the equation of a circle:
var radius = Math.pow((2*Math.pow(365, 2)), 1/2);
var t = 365; //offset
this.tMax = (Math.pow(Math.pow(r, 2) - Math.pow(x, 2), 1/2) - t) * (t / (r - t)) + 1;
the above equation has the above specified asymptotes. it is part of a step polynomial for scaling an axis for a flotr graph.
well, you are missing data (you need another point to determine the polynomial)
a*(x-1)^2+b*(x-1)+c=y-1
a*(x-365)^2+b*(x-365)+c=y-1
you can solve the exact answer for b
but A depends on C (or vv)
and your question is off topic anyways, and you need to revise your algebra
I know:
The control points a and d (start and end point of a 2D cubic bezier curve)
The slopes a->b, c->d, and b->c (b,c the other control points)
Where the halfway point of the Bézier curve is.
Now, given this information, what is the formula for the positions of control points b and c ?
I know this question is old, but there is no correct or complete answer provided, so I thought I'd chime in with a solution. Note that David's calculations contain several errors and his solution is incomplete even if these errors are corrected.
First, define vectors T0, T1 and T2 using the three slopes:
T0 = ( b - a ) / u0
T1 = ( c - b ) / u1
T2 = ( d - c ) / u2
If we knew both the direction and distance between each pair of control points then we would not need the scale factors u0, u1 and u2. Since we only know slope then u0, u1 and u2 are unknown scalar quantities. Also, we assume that u0, u1 and u2 are nonzero since slope is defined.
We can rewrite these equations in several different ways to obtain expressions for each control point in terms of the other control points. For example:
b = a + T0*u0
c = b + T1*u1
d = c + T2*u2
The question also states that we have the "halfway point" of the cubic Bezier curve. I take this to mean we have the point at the midpoint of the curve's parameter range. I will call this point p:
p = ( a + 3*b + 3*c + d ) / 8
Rewriting with unknowns on the left hand side yields:
b + c = ( 8*p - a - d ) / 3
We can now substitute for b and c in various ways using the earlier expressions. It turns out that ambiguities arise when we have parallel vectors T0, T1 or T2. There are four cases to consider.
Case 1: T0 is not parallel to T1
Substitute b = a + T0*u0 and c = a + T0*u0 + T1*u1 and solve for u0 and u1:
2*T0*u0 + T1*u1 = ( 8*p - 7*a - d ) / 3
This is two equations and two unknowns since T0 and T1 are vectors. Substitute u0 and u1 back into b = a + T0*u0 and c = a + T0*u0 + T1*u1 to obtain the missing control points b and c.
Case 2: T1 is not parallel to T2
Substitute c = d - T2*u2 and b = d - T2*u2 - T1*u1 and solve for u1 and u2:
T1*u1 + 2*T2*u2 = ( a + 7*d - 8*p ) / 3
Case 3: T0 is not parallel to T2
Substitute b = a + T0*u0 and c = d - T2*u2 and solve for u0 and u2:
T0*u0 - T2*u2 = ( 8*p - 4*a - 4*d ) / 3
Case 4: T0, T1 and T2 are all parallel
In this case a, b, c and d are all collinear and T0, T1 and T2 are all equivalent to within a scale factor. There is not enough information to obtain a unique solution. One simple solution would be to simply pick b by setting u0 = 1:
b = a + T0
(a + T0) + c = ( 8*p - a - d ) / 3
c = ( 8*p - 4*a - d - 3*T0 ) / 3
An infinite number of solutions exist. In essence, picking b defines c or picking c will define b.
Extending to 3D
The question specifically asked about planar Bezier curves, but I think it's interesting to note that the point p is not necessary when extending this problem to a non-planar 3D cubic Bezier curve. In this case, we can simply solve this equation for u0, u1 and u2:
T0*u0 + T1*u1 + T2*u2 = d - a
This is three equations (the vectors are 3D) and three unknowns (u0, u1 and u2). Substitution into b = a + T0*u0 and c = b + T1*u1 or c = d - T2*u2 yields b and c.
Let's say your slopes are normalized, then for some u,v you have
u * slope(a->b)+a = b, v * slope(c->d)+d = c
you know the values of a,d, and q:=(a+b+c+d)/8 (the halfway point of the curve)
so c = 8(q-a-d-b)
plugging the above equations in the last one you get
v * slope(c->d)+d = 8(q-a-d-a-u * slope(a->b))
which is 2 equations (a 2d vector equation) in two variables (u,v)
You don't need the third slope.