I *think* this is the answer you're looking for...

Create one matrix for the body orientation using the D3DXMatrixRotation... functions. Call this BodyRot.

Create another matrix for the eye orientation using the D3DX functions. Call this EyeRot.

Create another matrix with D3DXMatrixTranslation and use the parameters (0.0f, 0.0f, WalkingDistance). Call it WalkDistance.

Now, concatenate the matrices as follows (order is very important):

FinalMatrix = EyeRot * WalkDistance * BodyRot.

It makes sense if you read it backwards:

Rotate the body to a new direction. Walk X units along the new Z direction (changed by the rotation). Then, turn your head by a certain rotation. If you draw it out on paper (with transformed axes) it makes sense.

This is the disadvantage of using LookAt. LookAt is easier, and you can retrieve these components from the matrix, but this is a much better approach, especially if you ever get into the "face one way, but fire the gun in the walking direction" kind of thing.

Is that what you were looking for?

Edited by - CrazedGenius on January 15, 2002 3:00:51 PM

Rather then rotate your camera around the world basis, maintain a local basis around your camera. Then when the camera rotates you adjust the camera''s local basis accordingly. This also solves the gimbal lock problem. Gary Simmons had an article on this topic so your best bet is to head over to www.Mr-Gamemaker.com and post a message asking for the old tutorials. Also search for "Vector Camera".

ECKILLER

quote:Original post by FeelEvil
Translation: Next you need to calculate the translation matrix depending on the angles of the camera . If I just add to the z element, as people are telling me to do everywhere, the camera will only move in the z direction.

I think this is the point of confusion...

Get out a sheet of paper. Draw an X and Z axis (Y points out of the page). Draw a vector travelling straight down the Z axis. So far, so good.

Now, apply a rotation around the Y axis, say 45 degrees. Redraw the X and Z axes at this new 45 degree orientation. Redraw the vector straight down the Z axis. This vector should be 45 degrees from the old one, yet the translation (straight down Z) is the same in both cases. This is the direction of the body motion and of the eye vector.

Now, apply some final rotation, but no translation. This will affect the eye vector without "moving the body"

To put it another way, start over and draw the X and Z axis on the paper. Rotate 90 degrees. Now, after that rotation, a +Z translation will actually translate the object in either the -X or +X "universal" direction (depending on which direction you rotated.

As far as code, it's the same as I said before. A quick proof of this involves writing your paper experiment as code. Try different matrix multiplication orders, BUT USE THE EXACT SAME MATRICES (caps for emphasis, not yelling)

"Translate and then rotate" (R * T) = move to a new position and spin.

"Rotate and then translate" (T * R) = an orbitting effect because the object is rotated to a new angle and then moved along that angle.

Two different effects with the same matrices.

Please keep asking if it doesn't make sense.

Edited by - CrazedGenius on January 15, 2002 5:33:19 PM

This is taken directly from some of my old working DOS code. So, it's NOT neat but it will show you the math. Here ya go:

tempX = pntX;
pntX = (pntZ * sinDir) +
(pntX * cosDir);
pntZ = (pntZ * cosDir) -
(tempX * sinDir);

ltrans[_a].z = (pntY * sinPitch) +
(pntZ * cosPitch);
pntY = (pntY * cosPitch) -
(pntZ * sinPitch);

if (roll) {
ltrans[_a].x = (pntX * cosRoll) -
(pntY * sinRoll);
ltrans[_a].y = (pntX * sinRoll) +
(pntY * cosRoll);
}
else {
ltrans[_a].x = pntX;
ltrans[_a].y = pntY;
}

Man, that is hideous... but I'll try to explain. This is really just doing a 3D transformation based on Direction, Pitch and Roll... yes, I KNOW it's unorthadox... this was from my 1st attempt at a 3D engine years ago. ltrans is an array of points containing x, y and z coordinates and PntX, PntY and PntZ contain the temporary coordinates to be translated. The values, such as sinDir and cosDir, are double precision values containing the cosine and sin values for Dir, Pitch and Roll.

Now I know this isn't *exactly* what you are looking for but you can use this math to find the point that is ahead of the camera and move the camera to that point. In this case you could exclude the calculations for "roll" You should be able to transform the point (0, 0, x) where x is the distance you want the camera to move, and add that to your camera coordinates.

- Jay


Many of the truths we cling to depend greatly on our own point of view

Get Tranced!


Edited by - coderx75 on January 15, 2002 5:35:50 PM