FPS Camera

With Quaternions you can only represent rotations, so you must be using a separate vector for translations. Anyway, how do you interpolate rotation? Do you use slerp? Or nlerp? How do you interpolate the translation?

Quote:

Original post by dv With Quaternions you can only represent rotations, so you must be using a separate vector for translations. Anyway, how do you interpolate rotation? Do you use slerp? Or nlerp? How do you interpolate the translation?

Basically I neither slerp nor nlerp, why would I do that? I mean I could slerp and I do slerp in some cases, for example if I want to reset my camera to a starting rotation, i slerp to that rotation. But for movement I still use the 3 vectors, Look, Up, right, but at every step I calculate the quaternion rotation around these exes and then I make a matrix from that quaternion. Im not sure how I can slerp if its a first person camera and it moves with the mouse.

With quaternions i usually do something like this

// initQuaternion oldRotation = Identity();Quaternion newRotation = Identity();// stepMouseMovement mouse = grabLatestMouseMovement();oldRotation = newRotation;newRotation *= QuaternionFromAxisAngle(yPositive, mouse.dy * config.sensitivity_y * deltaT);newRotation *= QuaternionFromAxisAngle(oldRotation.getRight(), mouse.dx * config.sensitivity_x * deltaT);// renderorientCameraAcordingWithRotation( interpolate(oldRotation, deltaT_render, newRotation) );

where getRight() (and similar functions such as getUp() & getIn() roughly look like:

vec3 Quaternion::getRight() {  return rotatePointAroundOrigin(xPositive);}

hth

Quaternions are really overkill for a FPS camera. The mouse's vertical movement controls pitch, and its horizontal movement controls yaw, which makes Euler angles the most logical representation. You let YPRx (or xRPY) be your transformation order and expand out the matrix multiplication symbolically to get rid of multiplication by 0. Plug in yaw, pitch, roll, and the offset vector and you have the camera-space matrix. Invert using the special case method for homogeneous transformation matrices, and you have the matrix you pass to the API. All in all it goes something like this:



I made some liberties in the code just to demonstrate the process. You can actually eliminate the inverse completely if you set up the initial matrix correctly. Consider that a quaternion->matrix conversion is just as long, minus the offset part. And you still need to perform the original quaternion construction.

Quaternions and FPS cameras are a pet peeve of jyk so he'll probably add anything I've missed [grin]

Quote:

Original post by Zipster Quaternions are really overkill for a FPS camera. The mouse's vertical movement controls pitch, and its horizontal movement controls yaw, which makes Euler angles the most logical representation. You let YPRx (or xRPY) be your transformation order and expand out the matrix multiplication symbolically to get rid of multiplication by 0. Plug in yaw, pitch, roll, and the offset vector and you have the camera-space matrix. Invert using the special case method for homogeneous transformation matrices, and you have the matrix you pass to the API.

Ah, you beat me to it! I guess I'll just have to lie in wait for the next 'quaternion-based FPS camera' post (muhaha).

:)

K, thanks for responses guys