# point cloud alignment

This topic is 3412 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

## Recommended Posts

I'm trying to do basically do point cloud alignment, but aligning to a specific plane. I'm generating a heightmap form the points, but they're slightly sloped, with that i mean, the ground plane isn't on the x and y plane, but rather sloped up. so i want to align it back down, in order for the heightmap to not be rotated by a certain angle. i calculated my eigenvectors and ended up with -0.8703 -0.0393 -0.0905 0.0000 -0.3573 -0.6427 0.0002 -0.3027 -0.6972 but then i'm stuck, i don't know what to do next in order to do what i want. i've tried multiplying all 3d points with a matrix made up of the eigenvector, and that didn't work, i tried multiplying them by a 3x3 rotation matrix created by using the 1st and 2nd eigenvectors and this code,
float** MathUtil::calculateAlignmentRotMat(float* a, int al, float* b, int bl) {

float angle = acosf(dot3(a, b));
float* rotationAxis = cross3(a, b);

return calculateRotationMat(rotationAxis, angle);
}

float** MathUtil::calculateRotationMat(float* rotationAxis, float angle) {
float **result = new float*[3];

for (int i = 0; i < 3; i++)
result = new float[3];

float rx = rotationAxis[0];
float ry = rotationAxis[1];
float rz = rotationAxis[2];

delete[] rotationAxis;

float cosAngle = cosf(angle);
float sinAngle = sinf(angle);
float c = 1.0 - cosAngle;

result[0][0] = cosAngle + rx * rx * c;
result[1][0] = rz * sinAngle + ry * rx * c;
result[2][0] = -ry * sinAngle + rz * rx * c;
result[0][1] = -rz * sinAngle + rx * ry * c;
result[1][1] = cosAngle + ry * ry * c;
result[2][1] = rx * sinAngle + rz * ry * c;
result[0][2] = ry * sinAngle + rx * rz * c;
result[1][2] = -rx * sinAngle + ry * rz * c;
result[2][2] = cosAngle + rz * rz * c;
return result;
}

but that also didn't work. my general apply function is
    rotMatrx = util->calculateAlignmentRotMat(vecRot[0], m, vecRot[1], m);

float* interm2 = vector(m+1);
float* tempm = vector(m+1);

/* Form projections of row-points on first three prin. components. */
/* Store in 'realData', overwriting original data. */
for (i = 0; i < realN; i++) {
for (j = 0; j < m; j++) {
interm2[j] = realData[j];
}
interm2[3] = 1;
/* data[j] will be overwritten */
for (k = 0; k < 3; k++) {
for (k2 = 0; k2 < m; k2++) {
tempm[k2] += interm2[k2] * rotMatrx[k][k2];
}
}
realData[0] = tempm[0];
realData[1] = tempm[1];
realData[2] = tempm[2];

}

I really hope someone can help me out with this, Thanks Phyx

##### Share on other sites
I don't fully understand your question. Can you try to clarify exactly what it is you want?

##### Share on other sites
In a nutshell, this http://i43.tinypic.com/27xo6c5.png but in 3d.

##### Share on other sites
for(each point)
point.z = (constant)

Will that do?

##### Share on other sites
I don't think that'll work, mapping all the points to the same constant Z will in effect flatten the pointcloud, losing any and all structural information i had would it not. In the picture i had, the line is rather the direction of the point cloud. the after picture still has the points in the same location relative to this direction vector.

##### Share on other sites
Try computing the Least Squares Plane and then transform the points by the inverse transform of the plane.

##### Share on other sites
Ok so your diagram didn't show what you wanted :D

In that case you have no choice but to do a coordinate transformation as _swx_ suggests. Find the basis vectors and origin point for the plane that the point cloud is based on; this will give you a 4×4 transformation matrix m of the type:

( ....x.... 0 )( ....y.... 0 )( ....n.... 0 )( ....o.... 1 )

... where o is the origin point of the plane, n is the normal and x and y are two perpendicular basis vectors within the plane. (If your plane is only a little deviant from the target space, I recommend you make these vectors x=n×Y and y=X×n respectively, which will align them with the target's X and Y axes as closely as possible.)

Similarly, find the basis vectors for the plane onto which you which to project the point cloud, which make up a matrix M:
( ....X.... 0 )( ....Y.... 0 )( ....N.... 0 )( ....O.... 1 )

It is likely that your target axes X, Y and N will be the world axes [1,0,0], [0,1,0] and [0,0,1], making this a simple transformation matrix, but they do not have to be.

Now, to project a point p out of the coordinate space defined by m, and into the coordinate space defined by M, simply multiply by the inverse of m and then by M:

P = pm-1M

##### Share on other sites
Ah, ok, thanks _swx_ and Bob Janova , yeah i do beleive my world axes are [1,0,0], [0,1,0] and [0,0,1]. and i think i already have m, so but i was only applying it to my points, so that might be why it was screwed up. I'll try the one you mentioned. Thanks :)

1. 1
2. 2
Rutin
24
3. 3
4. 4
JoeJ
18
5. 5

• 14
• 22
• 11
• 11
• 9
• ### Forum Statistics

• Total Topics
631766
• Total Posts
3002218
×