Related
I'm drawing a blank at the moment. I need to translate back and forth between "signed" and "unsigned" degrees, [-180..180] and [0..360]
The simple way to go from [-180..180] to [0..360] is
(d+360) % 360 // (+360 removes ambiguity about the sign in some languages).
How do I do the inverse operation? I can do
if d<180 d else d-360
but it looks ugly.
Edit:
Here are some example numbers. I want to map this
[0, 90, 180, 270, 360]
to this:
[0, 90, 180, -90, 0]
Edit 2:
OK, brain freeze is over. The answer is
(a+180)%360-180
OK, brain freeze over the answer is
(a+180)%360-180
For mapping a interval to another interval, you can use a linear interpolation
like
float map(float n, float x1, float x2, float y1, float y2)
{
float m = (y2 - y1) / (x2 - x1);
return y1 + m * (n - x1);
}
where x1 <= n <= x2, results a value between y1 and y2
i'm programming a tree to divide the 2d-space.
therefore i need to split the current set of points in two equal parts according to a given angle. so the black line from the image attached below goes through the calculated median.
As we can see in the image, the line (defined by the given angle and the computed median) is far away from dividing the set in two equal parts. the line should cut the x-axis more to the right as now only ~25 points are on the left of the line and ~55 points on the right.
i already know, i have to use the angles of the point for my median computation.
my first approach for calculating the median was rotating all points by the given angle and then use their angles to the x-axis, but that's obviously not enough.
I'm not concretly stuck on the transformations i have to apply to the point in order to have a correct median computation, so any help/hint would be appreciated,
thanks a lot.
Normal form of line equation is
x * Cos(Theta) + y * Sin(Theta) - p = 0
where
Theta = Pi/2 + Fi
and Fi is your given angle.
You have to find appropriate p value to divide point set to two equal parts.
Signed distance from this line to point Q is
D = Q.X * Cos(Theta) + Q.Y * Sin(Theta) - p
Sort your point array by D[i] value and find median.
If point number is odd, get D[median] as p for line equation, otherwise you may choose any p value between D[leftmedian] and D[rightmedian]
Examples:
5 points, D = (-3, -1, 4, 5, 8) - get p = 4 (line passes through one point)
6 points, D = (-3, -1, 2, 5, 7, 9) - get any p in range ]2..5[ (not including these values)
Edit: two examples for odd and even point numbers:
code (Delphi):
var
Pt: TArray<TPoint>;
D: TArray<Double>;
N, i: Integer;
Fi, Theta, p: Double;
begin
Randomize;
N := 7 + Random(2); // number of points, 7 or 8
Fi := Pi / 4; //line slope angle
Theta := Pi / 2 + Fi;
SetLength(Pt, N); //point cloud
SetLength(D, N); //distance params
Canvas.FillRect(ClientRect);
//generate random points and draw them
for i := 0 to N - 1 do begin
Pt[i] := Point(Random(400), Random(400));
Canvas.Rectangle(Pt[i].X - 2, Pt[i].Y - 2, Pt[i].X + 3, Pt[i].Y + 3);
end;
//fill param array
for i := 0 to N - 1 do
D[i] := Pt[i].X * Cos(Theta) + Pt[i].Y * Sin(Theta);
//sort distance parameters
TArray.Sort<Double>(D);
//get line parameter as median of array
if Odd(N) then
p := D[N div 2]
else
p := (D[N div 2 - 1] + D[N div 2]) / 2;
//draw calculated line
Canvas.MoveTo(0, Round(p / Sin(Theta)));
Canvas.LineTo(500, Round((p - 500 * Cos(Theta)) / Sin(Theta)));
I've gotten stuck getting my euler angles out my rotation matrix.
My conventions are:
Left-handed (x right, z back, y up)
YZX
Left handed angle rotation
My rotation matrix is built up from Euler angles like (from my code):
var xRotationMatrix = $M([
[1, 0, 0, 0],
[0, cx, -sx, 0],
[0, sx, cx, 0],
[0, 0, 0, 1]
]);
var yRotationMatrix = $M([
[ cy, 0, sy, 0],
[ 0, 1, 0, 0],
[-sy, 0, cy, 0],
[ 0, 0, 0, 1]
]);
var zRotationMatrix = $M([
[cz, -sz, 0, 0],
[sz, cz, 0, 0],
[ 0, 0, 1, 0],
[ 0, 0, 0, 1]
]);
Which results in a final rotation matrix as:
R(YZX) = | cy.cz, -cy.sz.cx + sy.sx, cy.sz.sx + sy.cx, 0|
| sz, cz.cx, -cz.sx, 0|
|-sy.cz, sy.sz.cx + cy.sx, -sy.sz.sx + cy.cx, 0|
| 0, 0, 0, 1|
I'm calculating my euler angles back from this matrix using this code:
this.anglesFromMatrix = function(m) {
var y = 0, x = 0, z = 0;
if (m.e(2, 1) > 0.999) {
y = Math.atan2(m.e(1, 3), m.e(3, 3));
z = Math.PI / 2;
x = 0;
} else if (m.e(2, 1) < -0.999) {
y = Math.atan2(m.e(1, 3), m.e(3, 3));
z = -Math.PI / 2;
x = 0;
} else {
y = Math.atan2(-m.e(3, 1), -m.e(1, 1));
x = Math.atan2(-m.e(2, 3), m.e(2, 2));
z = Math.asin(m.e(2, 1));
}
return {theta: this.deg(x), phi: this.deg(y), psi: this.deg(z)};
};
I've done the maths backwards and forwards a few times, but I can't see what's wrong. Any help would hugely appreciated.
Your matrix and euler angles aren't consistent. It looks like you should be using
y = Math.atan2(-m.e(3, 1), m.e(1, 1));
instead of
y = Math.atan2(-m.e(3, 1), -m.e(1, 1));
for the general case (the else branch).
I said "looks like" because -- what language is this? I'm assuming you have the indexing correct for this language. Are you sure about atan2? There is no single convention for atan2. In some programming languages the sine term is the first argument, in others, the cosine term is the first argument.
The last and most important branch of the anglesFromMatrix function has a small sign error but otherwise works correctly. Use
y = Math.atan2(-m.e(3, 1), m.e(1, 1))
since only m.e(3, 1) of m.e(1, 1) = cy.cz and m.e(3, 1) = -sy.cz should be inverted. I haven't checked the other branches for errors.
Beware that since sz = m.e(2, 1) has two solutions, the angles (x, y, z) used to construct the matrix m might not be the same as the angles (rx, ry, rz) returned by anglesFromMatrix(m). Instead we can test that the matrix rm constructed from (rx, ry, rz) does indeed equal m.
I worked on this problem extensively to come up with the correct angles for a given matrix. The problem in the math comes from the inability to determine a precise value for the SIN since -SIN(x) = SIN(-x) and this will affect the other values of the matrix. The solution I came up with comes up with two equally valid solutions out of eight possible solutions. I used a standard Z . Y . X matrix form but it should be adaptable to any matrix. Start by findng the three angles from: X = atan(m32,m33): Y = -asin(m31) : Z = atan(m21,m11) : Then create angles X' = -sign(X)*PI+X : Y'= sign(Y)*PI-Y : Z = -sign(Z)*pi+Z . Using these angles create eight set of angle groups : XYZ : X'YZ : XYZ' : X'YZ' : X'Y'Z' : XY'Z' : X'Y'Z : XY'Z
Use these set to create the eight corresponding matrixes. Then do a sum of the difference between the unknown matrix and each matrix. This is a sum of each element of the unknown minus the same element of the test matrix. After doing this, two of the sums will be zero and those matrixes will represent the solution angles to the original matrix. This works for all possible angle combinations including 0's. As 0's are introduced, more of the eight test matrixes become valid. At 0,0,0 they all become idenity matrixes!
Hope this helps, it worked very well for my application.
Bruce
update
After finding problems with Y = -90 or 90 degrees in the solution above. I came up with this solution that seems to reproduce the matrix at all values!
X = if(or(m31=1,m31=-1),0,atan(m33+1e-24,m32))
Y = -asin(m31)
Z = if(or(m31=1,m31=-1),-atan2(m22,m12),atan2(m11+1e-24,m21))
I went the long way around to find this solution, but it wa very enlightening :o)
Hope this helps!
Bruce
I have a set of points. I want to separate them into 2 distinct sets. To do this, I choose two points (a and b) and draw an imaginary line between them. Now I want to have all points that are left from this line in one set and those that are right from this line in the other set.
How can I tell for any given point z whether it is in the left or in the right set? I tried to calculate the angle between a-z-b – angles smaller than 180 are on the right hand side, greater than 180 on the left hand side – but because of the definition of ArcCos, the calculated angles are always smaller than 180°. Is there a formula to calculate angles greater than 180° (or any other formula to chose right or left side)?
Try this code which makes use of a cross product:
public bool isLeft(Point a, Point b, Point c){
return ((b.X - a.X)*(c.Y - a.Y) - (b.Y - a.Y)*(c.X - a.X)) > 0;
}
Where a = line point 1; b = line point 2; c = point to check against.
If the formula is equal to 0, the points are colinear.
If the line is horizontal, then this returns true if the point is above the line.
Use the sign of the determinant of vectors (AB,AM), where M(X,Y) is the query point:
position = sign((Bx - Ax) * (Y - Ay) - (By - Ay) * (X - Ax))
It is 0 on the line, and +1 on one side, -1 on the other side.
You look at the sign of the determinant of
| x2-x1 x3-x1 |
| y2-y1 y3-y1 |
It will be positive for points on one side, and negative on the other (and zero for points on the line itself).
The vector (y1 - y2, x2 - x1) is perpendicular to the line, and always pointing right (or always pointing left, if you plane orientation is different from mine).
You can then compute the dot product of that vector and (x3 - x1, y3 - y1) to determine if the point lies on the same side of the line as the perpendicular vector (dot product > 0) or not.
Using the equation of the line ab, get the x-coordinate on the line at the same y-coordinate as the point to be sorted.
If point's x > line's x, the point is to the right of the line.
If point's
x < line's x, the point is to the left of the line.
If point's x == line's x, the point is on the line.
I implemented this in java and ran a unit test (source below). None of the above solutions work. This code passes the unit test. If anyone finds a unit test that does not pass, please let me know.
Code: NOTE: nearlyEqual(double,double) returns true if the two numbers are very close.
/*
* #return integer code for which side of the line ab c is on. 1 means
* left turn, -1 means right turn. Returns
* 0 if all three are on a line
*/
public static int findSide(
double ax, double ay,
double bx, double by,
double cx, double cy) {
if (nearlyEqual(bx-ax,0)) { // vertical line
if (cx < bx) {
return by > ay ? 1 : -1;
}
if (cx > bx) {
return by > ay ? -1 : 1;
}
return 0;
}
if (nearlyEqual(by-ay,0)) { // horizontal line
if (cy < by) {
return bx > ax ? -1 : 1;
}
if (cy > by) {
return bx > ax ? 1 : -1;
}
return 0;
}
double slope = (by - ay) / (bx - ax);
double yIntercept = ay - ax * slope;
double cSolution = (slope*cx) + yIntercept;
if (slope != 0) {
if (cy > cSolution) {
return bx > ax ? 1 : -1;
}
if (cy < cSolution) {
return bx > ax ? -1 : 1;
}
return 0;
}
return 0;
}
Here's the unit test:
#Test public void testFindSide() {
assertTrue("1", 1 == Utility.findSide(1, 0, 0, 0, -1, -1));
assertTrue("1.1", 1 == Utility.findSide(25, 0, 0, 0, -1, -14));
assertTrue("1.2", 1 == Utility.findSide(25, 20, 0, 20, -1, 6));
assertTrue("1.3", 1 == Utility.findSide(24, 20, -1, 20, -2, 6));
assertTrue("-1", -1 == Utility.findSide(1, 0, 0, 0, 1, 1));
assertTrue("-1.1", -1 == Utility.findSide(12, 0, 0, 0, 2, 1));
assertTrue("-1.2", -1 == Utility.findSide(-25, 0, 0, 0, -1, -14));
assertTrue("-1.3", -1 == Utility.findSide(1, 0.5, 0, 0, 1, 1));
assertTrue("2.1", -1 == Utility.findSide(0,5, 1,10, 10,20));
assertTrue("2.2", 1 == Utility.findSide(0,9.1, 1,10, 10,20));
assertTrue("2.3", -1 == Utility.findSide(0,5, 1,10, 20,10));
assertTrue("2.4", -1 == Utility.findSide(0,9.1, 1,10, 20,10));
assertTrue("vertical 1", 1 == Utility.findSide(1,1, 1,10, 0,0));
assertTrue("vertical 2", -1 == Utility.findSide(1,10, 1,1, 0,0));
assertTrue("vertical 3", -1 == Utility.findSide(1,1, 1,10, 5,0));
assertTrue("vertical 3", 1 == Utility.findSide(1,10, 1,1, 5,0));
assertTrue("horizontal 1", 1 == Utility.findSide(1,-1, 10,-1, 0,0));
assertTrue("horizontal 2", -1 == Utility.findSide(10,-1, 1,-1, 0,0));
assertTrue("horizontal 3", -1 == Utility.findSide(1,-1, 10,-1, 0,-9));
assertTrue("horizontal 4", 1 == Utility.findSide(10,-1, 1,-1, 0,-9));
assertTrue("positive slope 1", 1 == Utility.findSide(0,0, 10,10, 1,2));
assertTrue("positive slope 2", -1 == Utility.findSide(10,10, 0,0, 1,2));
assertTrue("positive slope 3", -1 == Utility.findSide(0,0, 10,10, 1,0));
assertTrue("positive slope 4", 1 == Utility.findSide(10,10, 0,0, 1,0));
assertTrue("negative slope 1", -1 == Utility.findSide(0,0, -10,10, 1,2));
assertTrue("negative slope 2", -1 == Utility.findSide(0,0, -10,10, 1,2));
assertTrue("negative slope 3", 1 == Utility.findSide(0,0, -10,10, -1,-2));
assertTrue("negative slope 4", -1 == Utility.findSide(-10,10, 0,0, -1,-2));
assertTrue("0", 0 == Utility.findSide(1, 0, 0, 0, -1, 0));
assertTrue("1", 0 == Utility.findSide(0,0, 0, 0, 0, 0));
assertTrue("2", 0 == Utility.findSide(0,0, 0,1, 0,2));
assertTrue("3", 0 == Utility.findSide(0,0, 2,0, 1,0));
assertTrue("4", 0 == Utility.findSide(1, -2, 0, 0, -1, 2));
}
First check if you have a vertical line:
if (x2-x1) == 0
if x3 < x2
it's on the left
if x3 > x2
it's on the right
else
it's on the line
Then, calculate the slope: m = (y2-y1)/(x2-x1)
Then, create an equation of the line using point slope form: y - y1 = m*(x-x1) + y1. For the sake of my explanation, simplify it to slope-intercept form (not necessary in your algorithm): y = mx+b.
Now plug in (x3, y3) for x and y. Here is some pseudocode detailing what should happen:
if m > 0
if y3 > m*x3 + b
it's on the left
else if y3 < m*x3 + b
it's on the right
else
it's on the line
else if m < 0
if y3 < m*x3 + b
it's on the left
if y3 > m*x3+b
it's on the right
else
it's on the line
else
horizontal line; up to you what you do
I wanted to provide with a solution inspired by physics.
Imagine a force applied along the line and you are measuring the torque of the force about the point. If the torque is positive (counterclockwise) then the point is to the "left" of the line, but if the torque is negative the point is the "right" of the line.
So if the force vector equals the span of the two points defining the line
fx = x_2 - x_1
fy = y_2 - y_1
you test for the side of a point (px,py) based on the sign of the following test
var torque = fx*(py-y_1)-fy*(px-x_1)
if torque>0 then
"point on left side"
else if torque <0 then
"point on right side"
else
"point on line"
end if
Assuming the points are (Ax,Ay) (Bx,By) and (Cx,Cy), you need to compute:
(Bx - Ax) * (Cy - Ay) - (By - Ay) * (Cx - Ax)
This will equal zero if the point C is on the line formed by points A and B, and will have a different sign depending on the side. Which side this is depends on the orientation of your (x,y) coordinates, but you can plug test values for A,B and C into this formula to determine whether negative values are to the left or to the right.
basically, I think that there is a solution which is much easier and straight forward, for any given polygon, lets say consist of four vertices(p1,p2,p3,p4), find the two extreme opposite vertices in the polygon, in another words, find the for example the most top left vertex (lets say p1) and the opposite vertex which is located at most bottom right (lets say ). Hence, given your testing point C(x,y), now you have to make double check between C and p1 and C and p4:
if cx > p1x AND cy > p1y ==> means that C is lower and to right of p1
next
if cx < p2x AND cy < p2y ==> means that C is upper and to left of p4
conclusion, C is inside the rectangle.
Thanks :)
#AVB's answer in ruby
det = Matrix[
[(x2 - x1), (x3 - x1)],
[(y2 - y1), (y3 - y1)]
].determinant
If det is positive its above, if negative its below. If 0, its on the line.
Here's a version, again using the cross product logic, written in Clojure.
(defn is-left? [line point]
(let [[[x1 y1] [x2 y2]] (sort line)
[x-pt y-pt] point]
(> (* (- x2 x1) (- y-pt y1)) (* (- y2 y1) (- x-pt x1)))))
Example usage:
(is-left? [[-3 -1] [3 1]] [0 10])
true
Which is to say that the point (0, 10) is to the left of the line determined by (-3, -1) and (3, 1).
NOTE: This implementation solves a problem that none of the others (so far) does! Order matters when giving the points that determine the line. I.e., it's a "directed line", in a certain sense. So with the above code, this invocation also produces the result of true:
(is-left? [[3 1] [-3 -1]] [0 10])
true
That's because of this snippet of code:
(sort line)
Finally, as with the other cross product based solutions, this solution returns a boolean, and does not give a third result for collinearity. But it will give a result that makes sense, e.g.:
(is-left? [[1 1] [3 1]] [10 1])
false
Issues with the existing solution:
While I found Eric Bainville's answer to be correct, I found it entirely inadequate to comprehend:
How can two vectors have a determinant? I thought that applied to matrices?
What is sign?
How do I convert two vectors into a matrix?
position = sign((Bx - Ax) * (Y - Ay) - (By - Ay) * (X - Ax))
What is Bx?
What is Y? Isn't Y meant to be a Vector, rather than a scalar?
Why is the solution correct - what is the reasoning behind it?
Moreover, my use case involved complex curves rather than a simple line, hence it requires a little re-jigging:
Reconstituted Answer
Point a = new Point3d(ax, ay, az); // point on line
Point b = new Point3d(bx, by, bz); // point on line
If you want to see whether your points are above/below a curve, then you would need to get the first derivative of the particular curve you are interested in - also known as the tangent to the point on the curve. If you can do so, then you can highlight your points of interest. Of course, if your curve is a line, then you just need the point of interest without the tangent. The tangent IS the line.
Vector3d lineVector = curve.GetFirstDerivative(a); // where "a" is a point on the curve. You may derive point b with a simple displacement calculation:
Point3d b = new Point3d(a.X, a.Y, a.Z).TransformBy(
Matrix3d.Displacement(curve.GetFirstDerivative(a))
);
Point m = new Point3d(mx, my, mz) // the point you are interested in.
The Solution:
return (b.X - a.X) * (m.Y - a.Y) - (b.Y - a.Y) * (m.X - a.X) < 0; // the answer
Works for me! See the proof in the photo above. Green bricks satisfy the condition, but the bricks outside were filtered out! In my use case - I only want the bricks that are touching the circle.
Theory behind the answer
I will return to explain this. Someday. Somehow...
An alternative way of getting a feel of solutions provided by netters is to understand a little geometry implications.
Let pqr=[P,Q,R] are points that forms a plane that is divided into 2 sides by line [P,R]. We are to find out if two points on pqr plane, A,B, are on the same side.
Any point T on pqr plane can be represented with 2 vectors: v = P-Q and u = R-Q, as:
T' = T-Q = i * v + j * u
Now the geometry implications:
i+j =1: T on pr line
i+j <1: T on Sq
i+j >1: T on Snq
i+j =0: T = Q
i+j <0: T on Sq and beyond Q.
i+j: <0 0 <1 =1 >1
---------Q------[PR]--------- <== this is PQR plane
^
pr line
In general,
i+j is a measure of how far T is away from Q or line [P,R], and
the sign of i+j-1 implicates T's sideness.
The other geometry significances of i and j (not related to this solution) are:
i,j are the scalars for T in a new coordinate system where v,u are the new axes and Q is the new origin;
i, j can be seen as pulling force for P,R, respectively. The larger i, the farther T is away from R (larger pull from P).
The value of i,j can be obtained by solving the equations:
i*vx + j*ux = T'x
i*vy + j*uy = T'y
i*vz + j*uz = T'z
So we are given 2 points, A,B on the plane:
A = a1 * v + a2 * u
B = b1 * v + b2 * u
If A,B are on the same side, this will be true:
sign(a1+a2-1) = sign(b1+b2-1)
Note that this applies also to the question: Are A,B in the same side of plane [P,Q,R], in which:
T = i * P + j * Q + k * R
and i+j+k=1 implies that T is on the plane [P,Q,R] and the sign of i+j+k-1 implies its sideness. From this we have:
A = a1 * P + a2 * Q + a3 * R
B = b1 * P + b2 * Q + b3 * R
and A,B are on the same side of plane [P,Q,R] if
sign(a1+a2+a3-1) = sign(b1+b2+b3-1)
equation of line is y-y1 = m(x-x1)
here m is y2-y1 / x2-x1
now put m in equation and put condition on y < m(x-x1) + y1 then it is left side point
eg.
for i in rows:
for j in cols:
if j>m(i-a)+b:
image[i][j]=0
A(x1,y1) B(x2,y2) a line segment with length L=sqrt( (y2-y1)^2 + (x2-x1)^2 )
and a point M(x,y)
making a transformation of coordinates in order to be the point A the new start and B a point of the new X axis
we have the new coordinates of the point M
which are
newX = ((x-x1)(x2-x1)+(y-y1)(y2-y1)) / L
from (x-x1)*cos(t)+(y-y1)*sin(t) where cos(t)=(x2-x1)/L, sin(t)=(y2-y1)/L
newY = ((y-y1)(x2-x1)-(x-x1)(y2-y1)) / L
from (y-y1)*cos(t)-(x-x1)*sin(t)
because "left" is the side of axis X where the Y is positive, if the newY (which is the distance of M from AB) is positive, then it is on the left side of AB (the new X axis)
You may omit the division by L (allways positive), if you only want the sign
I have two vectors u and v. Is there a way of finding a quaternion representing the rotation from u to v?
Quaternion q;
vector a = crossproduct(v1, v2);
q.xyz = a;
q.w = sqrt((v1.Length ^ 2) * (v2.Length ^ 2)) + dotproduct(v1, v2);
Don't forget to normalize q.
Richard is right about there not being a unique rotation, but the above should give the "shortest arc," which is probably what you need.
Half-Way Vector Solution
I came up with the solution that I believe Imbrondir was trying to present (albeit with a minor mistake, which was probably why sinisterchipmunk had trouble verifying it).
Given that we can construct a quaternion representing a rotation around an axis like so:
q.w == cos(angle / 2)
q.x == sin(angle / 2) * axis.x
q.y == sin(angle / 2) * axis.y
q.z == sin(angle / 2) * axis.z
And that the dot and cross product of two normalized vectors are:
dot == cos(theta)
cross.x == sin(theta) * perpendicular.x
cross.y == sin(theta) * perpendicular.y
cross.z == sin(theta) * perpendicular.z
Seeing as a rotation from u to v can be achieved by rotating by theta (the angle between the vectors) around the perpendicular vector, it looks as though we can directly construct a quaternion representing such a rotation from the results of the dot and cross products; however, as it stands, theta = angle / 2, which means that doing so would result in twice the desired rotation.
One solution is to compute a vector half-way between u and v, and use the dot and cross product of u and the half-way vector to construct a quaternion representing a rotation of twice the angle between u and the half-way vector, which takes us all the way to v!
There is a special case, where u == -v and a unique half-way vector becomes impossible to calculate. This is expected, given the infinitely many "shortest arc" rotations which can take us from u to v, and we must simply rotate by 180 degrees around any vector orthogonal to u (or v) as our special-case solution. This is done by taking the normalized cross product of u with any other vector not parallel to u.
Pseudo code follows (obviously, in reality the special case would have to account for floating point inaccuracies -- probably by checking the dot products against some threshold rather than an absolute value).
Also note that there is no special case when u == v (the identity quaternion is produced -- check and see for yourself).
// N.B. the arguments are _not_ axis and angle, but rather the
// raw scalar-vector components.
Quaternion(float w, Vector3 xyz);
Quaternion get_rotation_between(Vector3 u, Vector3 v)
{
// It is important that the inputs are of equal length when
// calculating the half-way vector.
u = normalized(u);
v = normalized(v);
// Unfortunately, we have to check for when u == -v, as u + v
// in this case will be (0, 0, 0), which cannot be normalized.
if (u == -v)
{
// 180 degree rotation around any orthogonal vector
return Quaternion(0, normalized(orthogonal(u)));
}
Vector3 half = normalized(u + v);
return Quaternion(dot(u, half), cross(u, half));
}
The orthogonal function returns any vector orthogonal to the given vector. This implementation uses the cross product with the most orthogonal basis vector.
Vector3 orthogonal(Vector3 v)
{
float x = abs(v.x);
float y = abs(v.y);
float z = abs(v.z);
Vector3 other = x < y ? (x < z ? X_AXIS : Z_AXIS) : (y < z ? Y_AXIS : Z_AXIS);
return cross(v, other);
}
Half-Way Quaternion Solution
This is actually the solution presented in the accepted answer, and it seems to be marginally faster than the half-way vector solution (~20% faster by my measurements, though don't take my word for it). I'm adding it here in case others like myself are interested in an explanation.
Essentially, instead of calculating a quaternion using a half-way vector, you can calculate the quaternion which results in twice the required rotation (as detailed in the other solution), and find the quaternion half-way between that and zero degrees.
As I explained before, the quaternion for double the required rotation is:
q.w == dot(u, v)
q.xyz == cross(u, v)
And the quaternion for zero rotation is:
q.w == 1
q.xyz == (0, 0, 0)
Calculating the half-way quaternion is simply a matter of summing the quaternions and normalizing the result, just like with vectors. However, as is also the case with vectors, the quaternions must have the same magnitude, otherwise the result will be skewed towards the quaternion with the larger magnitude.
A quaternion constructed from the dot and cross product of two vectors will have the same magnitude as those products: length(u) * length(v). Rather than dividing all four components by this factor, we can instead scale up the identity quaternion. And if you were wondering why the accepted answer seemingly complicates matters by using sqrt(length(u) ^ 2 * length(v) ^ 2), it's because the squared length of a vector is quicker to calculate than the length, so we can save one sqrt calculation. The result is:
q.w = dot(u, v) + sqrt(length_2(u) * length_2(v))
q.xyz = cross(u, v)
And then normalize the result. Pseudo code follows:
Quaternion get_rotation_between(Vector3 u, Vector3 v)
{
float k_cos_theta = dot(u, v);
float k = sqrt(length_2(u) * length_2(v));
if (k_cos_theta / k == -1)
{
// 180 degree rotation around any orthogonal vector
return Quaternion(0, normalized(orthogonal(u)));
}
return normalized(Quaternion(k_cos_theta + k, cross(u, v)));
}
The problem as stated is not well-defined: there is not a unique rotation for a given pair of vectors. Consider the case, for example, where u = <1, 0, 0> and v = <0, 1, 0>. One rotation from u to v would be a pi / 2 rotation around the z-axis. Another rotation from u to v would be a pi rotation around the vector <1, 1, 0>.
I'm not much good on Quaternion. However I struggled for hours on this, and could not make Polaris878 solution work. I've tried pre-normalizing v1 and v2. Normalizing q. Normalizing q.xyz. Yet still I don't get it. The result still didn't give me the right result.
In the end though I found a solution that did. If it helps anyone else, here's my working (python) code:
def diffVectors(v1, v2):
""" Get rotation Quaternion between 2 vectors """
v1.normalize(), v2.normalize()
v = v1+v2
v.normalize()
angle = v.dot(v2)
axis = v.cross(v2)
return Quaternion( angle, *axis )
A special case must be made if v1 and v2 are paralell like v1 == v2 or v1 == -v2 (with some tolerance), where I believe the solutions should be Quaternion(1, 0,0,0) (no rotation) or Quaternion(0, *v1) (180 degree rotation)
Why not represent the vector using pure quaternions? It's better if you normalize them first perhaps.
q1 = (0 ux uy uz)'
q2 = (0 vx vy vz)'
q1 qrot = q2
Pre-multiply with q1-1
qrot = q1-1 q2
where q1-1 = q1conj / qnorm
This is can be thought of as "left division".
Right division, which is not what you want is:
qrot,right = q2-1 q1
From algorithm point of view , the fastest solution looks in pseudocode
Quaternion shortest_arc(const vector3& v1, const vector3& v2 )
{
// input vectors NOT unit
Quaternion q( cross(v1, v2), dot(v1, v2) );
// reducing to half angle
q.w += q.magnitude(); // 4 multiplication instead of 6 and more numerical stable
// handling close to 180 degree case
//... code skipped
return q.normalized(); // normalize if you need UNIT quaternion
}
Be sure that you need unit quaternions (usualy, it is required for interpolation).
NOTE:
Nonunit quaternions can be used with some operations faster than unit.
Some of the answers don't seem to consider possibility that cross product could be 0. Below snippet uses angle-axis representation:
//v1, v2 are assumed to be normalized
Vector3 axis = v1.cross(v2);
if (axis == Vector3::Zero())
axis = up();
else
axis = axis.normalized();
return toQuaternion(axis, ang);
The toQuaternion can be implemented as follows:
static Quaternion toQuaternion(const Vector3& axis, float angle)
{
auto s = std::sin(angle / 2);
auto u = axis.normalized();
return Quaternion(std::cos(angle / 2), u.x() * s, u.y() * s, u.z() * s);
}
If you are using Eigen library, you can also just do:
Quaternion::FromTwoVectors(from, to)
Working just with normalized quaternions, we can express Joseph Thompson's answer in the follwing terms.
Let q_v = (0, u_x, v_y, v_z) and q_w = (0, v_x, v_y, v_z) and consider
q = q_v * q_w = (-u dot v, u x v).
So representing q as q(q_0, q_1, q_2, q_3) we have
q_r = (1 - q_0, q_1, q_2, q_3).normalize()
According to the derivation of the quaternion rotation between two angles, one can rotate a vector u to vector v with
function fromVectors(u, v) {
d = dot(u, v)
w = cross(u, v)
return Quaternion(d + sqrt(d * d + dot(w, w)), w).normalize()
}
If it is known that the vectors u to vector v are unit vectors, the function reduces to
function fromUnitVectors(u, v) {
return Quaternion(1 + dot(u, v), cross(u, v)).normalize()
}
Depending on your use-case, handling the cases when the dot product is 1 (parallel vectors) and -1 (vectors pointing in opposite directions) may be needed.
The Generalized Solution
function align(Q, u, v)
U = quat(0, ux, uy, uz)
V = quat(0, vx, vy, vz)
return normalize(length(U*V)*Q - V*Q*U)
To find the quaternion of smallest rotation which rotate u to v, use
align(quat(1, 0, 0, 0), u, v)
Why This Generalization?
R is the quaternion closest to Q which will rotate u to v. More importantly, R is the quaternion closest to Q whose local u direction points in same direction as v.
This can be used to give you all possible rotations which rotate from u to v, depending on the choice of Q. If you want the minimal rotation from u to v, as the other solutions give, use Q = quat(1, 0, 0, 0).
Most commonly, I find that the real operation you want to do is a general alignment of one axis with another.
// If you find yourself often doing something like
quatFromTo(toWorldSpace(Q, localFrom), worldTo)*Q
// you should instead consider doing
align(Q, localFrom, worldTo)
Example
Say you want the quaternion Y which only represents Q's yaw, the pure rotation about the y axis. We can compute Y with the following.
Y = align(quat(Qw, Qx, Qy, Qz), vec(0, 1, 0), vec(0, 1, 0))
// simplifies to
Y = normalize(quat(Qw, 0, Qy, 0))
Alignment as a 4x4 Projection Matrix
If you want to perform the same alignment operation repeatedly, because this operation is the same as the projection of a quaternion onto a 2D plane embedded in 4D space, we can represent this operation as the multiplication with 4x4 projection matrix, A*Q.
I = mat4(
1, 0, 0, 0,
0, 1, 0, 0,
0, 0, 1, 0,
0, 0, 0, 1)
A = I - leftQ(V)*rightQ(U)/length(U*V)
// which expands to
A = mat4(
1 + ux*vx + uy*vy + uz*vz, uy*vz - uz*vy, uz*vx - ux*vz, ux*vy - uy*vx,
uy*vz - uz*vy, 1 + ux*vx - uy*vy - uz*vz, uy*vx + ux*vy, uz*vx + ux*vz,
uz*vx - ux*vz, uy*vx + ux*vy, 1 - ux*vx + uy*vy - uz*vz, uz*vy + uy*vz,
ux*vy - uy*vx, uz*vx + ux*vz, uz*vy + uy*vz, 1 - ux*vx - uy*vy + uz*vz)
// A can be applied to Q with the usual matrix-vector multiplication
R = normalize(A*Q)
//LeftQ is a 4x4 matrix which represents the multiplication on the left
//RightQ is a 4x4 matrix which represents the multiplication on the Right
LeftQ(w, x, y, z) = mat4(
w, -x, -y, -z,
x, w, -z, y,
y, z, w, -x,
z, -y, x, w)
RightQ(w, x, y, z) = mat4(
w, -x, -y, -z,
x, w, z, -y,
y, -z, w, x,
z, y, -x, w)