Camera 2 |
Top Previous Next |

1. By default, in OpenGL your viewpoint should be at (0,0,0) but looking down the -Z axis toward (0,0,-1), not (0,0,1).
2. Second, although gluLookAt() is very powerful and useful, you may find that you want to end up moving & aiming the camera yourself using calls to glTranslatef() and glRotatef() instead.
3. Make sure you're setting up your projection viewmatrix correctly. (Are you already seeing objects in 3d space okay?) You'll probably want to use gluPerspective() for this.
4. As far as gluLookAt() goes, here's the basic rundown:
gluLookAt( GLdouble eyex, // x-coordinate of camera's position GLdouble eyey, // y-coordinate of camera's position GLdouble eyez, // z-coordinate of camera's position GLdouble centerx, // x-coordinate of look-at point in 3d space GLdouble centery, // y-coordinate of look-at point in 3d space GLdouble centerz, // z-coordinate of look-at point in 3d space GLdouble upx, // x-component of a vector pointing "up" on the screen GLdouble upy, // y-component of a vector pointing "up" on the screen GLdouble upz, // z-component of a vector pointing "up" on the screen );
The <eye> and <center> triples are relatively self-explanatory (though ask if you have questions about them).
The one that's probably most confusing is the <up> vector. Consider this: by providing two points in space, <eye> and <center>, we establish (A) where the camera is at and (b) at point in space at which the camera is aimed. Now imagine a straight line drawn in 3d space going from point A to point B. We have NOT specified whether the camera is "right-side up", "upside-down", or any other weird and arbitrary rotation (roll) around this line. In order to do this, we specify an <up> vector for the camera. Essentially, we are saying, "if there were an antenna pointing out of the top of the camera, this is the direction it would be pointing".
As for your example (moving and turning on the XZ plane, with +Y as up): If you weren't able to turn around, the problem would be a simple one. Forward movement would simply mean "make the Z component of <eye> and <center> smaller (more negative)". Backward movement means "make <eyez> & <centerz> bigger <more positive>". Strafing left means "make <eyex> and <centerx> more negative"; strafing right means "make <eyex> and <centerx> more positive". (Modifying <eyey> and <centery> would make the character rise and fall vertically.)
Last but not least, your <up> vector would ALWAYS be (0,1,0) (i.e. the camera is "right-side up", antenna pointing in the +Y direction).
However, when turning becomes involved in a Quake-style fashion (which is how I read what you're describing), using gluLookAt() actually becomes a hassle. If instead we use glTranslatef() and glRotatef(), we can get a simpler solution.
Let's say we have the following information for the player / camera. I'm not sure if you're in C or C++ so we'll stick to floats:
float position[ 3 ]; // eye position in 3d space (0=x, 1=y, 2=z) float yawDegrees; // degrees of rotation, counter-clockwise around +Y. float pitchDegrees; // degrees of bending-over ("pitch" angle)
...and we have a macro (or, preferably, an inline function) to convert from degrees to radians:
#define PI 3.1415926535897932384626433832795 #define DEG2RAD( x ) ((x) * PI / 180.0)
When the camera moves forward <dist> units: position[ 0 ] -= dist * cos( DEG2RAD( yawDegrees ) ); position[ 2 ] -= dist * sin( DEG2RAD( yawDegrees ) );
When the camera strafes to the left <dist> units: position[ 0 ] -= dist * cos( DEG2RAD( yawDegrees + 90 ) ); position[ 2 ] -= dist * sin( DEG2RAD( yawDegrees + 90 ) );
When the camera moves back <dist> units: position[ 0 ] -= dist * cos( DEG2RAD( yawDegrees + 180 ) ); position[ 2 ] -= dist * sin( DEG2RAD( yawDegrees + 180 ) );
When the camera strafes to the right <dist> units: position[ 0 ] -= dist * cos( DEG2RAD( yawDegrees + 270 ) ); position[ 2 ] -= dist * sin( DEG2RAD( yawDegrees + 270 ) );
When the camera moves straight up <dist> units: position[ 1 ] += dist;
When the camera moves straight down <dist> units: position[ 1 ] -= dist;
When the camera turns to the right <turnAngle> degrees: yawDegrees -= turnAngle;
When the camera turns to the left <turnAngle> degrees: yawDegrees += turnAngle;
When the player looks up <pitchAngle> degrees: pitchDegrees += pitchAngle;
When the player looks down <pitchAngle> degrees: pitchDegrees -= pitchAngle;
Then, when it comes time to draw your scene, place the OpenGL camera like such:
/// Clear the camera's translation from the previous render frame glLoadIdentity();
/// Set up the perspective view gluPerspective( ...... ) // you should already have a line like this in your code
/// Move the camera to the player's position glTranslatef( -position[0], -position[1], -position[2] );
/// Rotate the view by <yawDegrees> around the Y-axis: glRotatef( yawDegrees, 0.0f, 1.0f, 0.0f );
/// Rotate the view by <pitchDegrees> around the new, rotated X-axis: glRotatef( pitchDegrees, 1.0f, 0.0f, 0.0f );
/// Now draw your objects in 3d space.
Note that we had to use DEG2RAD to convert our angle from degrees (0-360) to radians (0-2xPI), since the cos() and sin() functions take angles in radians. However, we did not have to do this for our glRotatef() command, since it takes angles in degrees.
I'm sure this is probably confusing and/or contains errors, but it's 2:20am so I claim no responsibility whatsoever. ;-)
Feel free to ask for clarification on any points that are confusing... admittedly, this is a poor overview of the vast subject of 3d view transformations, which deserves 20 pages of text rather than 2. In a large-scale 3d game - or one involving complex cameras and/or cameras floating in free space (with no discernible "up" or "down") - there are actually much better ways of treating angles than with yaw/pitch/roll (called Euler angles), namely Quaternions and Matricies.
cheers, -sq
p.s. Come to think of it, things might not be all that bad using gluLookAt() after all. We'd basically just have a 3d "forward" vector for the player (which we'd update whenever yaw-ing or pitch-ing occurred) and then set a view target at an arbitrary remote point along that vector. However, it's still probably conceptually simpler to think in terms of Euler angles for the time being. |