Hello,

I am hoping someone expert at matrix transformations can help me. I have an openGL app where I have a scene set up, and I allow the user to 'twist' rotate around a certain point in the scene. I use these calls to do it:


glPushMatrix();

glLoadMatrixf((GLfloat *)currentRotationMatrix);

glScalef(defaultWorldScale,defaultWorldScale,defaultWorldScale);
glScalef(drawScale,drawScale,drawScale);
glTranslatef(xTrans, yTrans, zTrans);
glTranslatef(twistx, twisty, twistz);

glGetFloatv(GL_MODELVIEW_MATRIX, TempRotationMatrix);
glRotatef(-zr, ((float *)TempRotationMatrix)[2], ((float *)TempRotationMatrix)[6], ((float *)TempRotationMatrix)[10]);

glTranslatef(-twistx, -twisty, -twistz);
glTranslatef(-xTrans, -yTrans, -zTrans);
glScalef(1.0/drawScale,1.0/drawScale,1.0/drawScale);
glScalef(1.0/defaultWorldScale,1.0/defaultWorldScale,1.0/defaultWorldScale);


glGetFloatv(GL_MODELVIEW_MATRIX, (GLfloat *)currentRotationMatrix); // save for later

glPopMatrix();

So, I have a currentRotationMatrix and then the other transforms are to get it out to a point in space where I want to do the twist. Then I do it and reverse my setup transformations.

This works, but what I end up with is a rotationMatrix that has been dirtied with extra translation values. I want to keep my rotationMatrix as only rotations. Is there any way to take the translation values that appear in my final matrix and incorporate them into my translation values xTrans yTrans zTrans? I have experimented but cannot figure out how to do it.

I know that the bottom left three values are translation, but I can't seem to get them separated so that I can just incorporate them into my main three translation values. For example, I would have thought I could take them and divide by (defaultWorldScale*drawScale), but it doesn't work.

I hope this question makes sense -- basically I try to compose my scene from xyz translation variables and XYZ euler rotation variables. But when I interactively make changes, I don't quite know how to separate them back out.

Thanks for any advice
Bob

Hello, thanks for your reply.

I don't really have "any number" of transformations. As you can see, I have a two scale values and two sets of translations. I do these, then perform my rotation, and then UNDO the first transformations. So that what I am left with is a rotation around some point in space.

If I do as you say and perform the final matrix on my origin, then I just get the translation values of the final matrix -- ie, the three values along the bottom left row of the matrix.

But I want to know how much I need to change xTrans, yTrans, and zTrans above so that I get the same final result. I guess I should show that this is how I normally put these values together, every frame:


glMultMatrixf((GLfloat*)currentRotationMatrix);

glScalef(defaultWorldScale,defaultWorldScale,defaultWorldScale);
glScalef(drawScale,drawScale,drawScale);

glTranslatef(xTrans, yTrans, zTrans);


So normally I have a rotation-only matrix, two scales, and a set of XYZ translations. But then interactively I am rotating the model as in my first post, and I end up with a currentRotationMatrix that has the correct rotation values but that also has some extra translations. I want to extract those back to my xTrans, yTrans, zTrans. I don't really understand why it isn't just a scale by my scale values, but that doesn't seem to work.

Bob

I got this to work. It was almost what I was assuming, that the resultant translation values would just need to be scaled by the reverse of my original scales. What I was missing was that the translation values in the final matrix were actually rotated by the matrix's own rotation. I needed to undo that first.

So I had to zero out the translation values, invert the matrix, and rotate those values using the invert:

sgVec4 p,w;

float sfactor = 1.0/(defaultWorldScale*drawScale);

p[0] = currentRotationMatrix[12]; // translation values of final 'dirty' rotation matrix
p[1] = currentRotationMatrix[13];
p[2] = currentRotationMatrix[14];

currentRotationMatrix[12] = currentRotationMatrix[13] = currentRotationMatrix[14] = 0.0; // set them to zero

OpenGL_Invert((GLfloat *)currentRotationMatrix, (GLfloat *)currentRotationMatrixInvert);

ConcatVector(p,(GLfloat *)currentRotationMatrixInvert, w);

xTrans += w[0]*sfactor;
yTrans += w[1]*sfactor;
zTrans += w[2]*sfactor;

Bob