I have wasted the better part of the last couple of days trying to
figure this out and still haven't found a satisfactory solution.
I'm trying to implement a mouse controlled rotating camera like the one
in Google Earth. Pick a point on the globe and drag it to rotate the
camera around. The mouse cursor should stay put on the anchor point. You
should be able to freely rotate around the globe - the poles should not
represent a special case.
I think the algorithm should look something like this:
1) find the anchor point on the globe on mouseDown
2) find the current cursor position on the globe on mouseMove
3) calculate the angular difference between 1) and 2), i.e. get a
rotation axis via cross product and an angle via dot product
4) rotate the camera accordingly
What I've tried so far (with various degrees of success):
- hovercamera. Won't allow tilting across the poles. Handles weirdly at
the poles. I think the rotational model of pan/tilt is too simple for
this case.
- rotate a targetCamera using matrix math. I basically ran into the same
problems as the hovercamera.
- rotate a targetCamera using quaternions. The most promising approach
so far. Works most of the time but I still have some unpredictable
"wobbles" when moving the mouse in a straight line (in screenspace).
- create a ray via camera.unproject of the current mouse position and
the camera position and intersect that with the sphere. I had numerical
instabilities with this approach which lead to jumpy movement.
- use a virtual trackball sphere at the origin and assume the camera is
always looking right at it. Works better than the ray/sphere
intersection tried before but now I'm missing aspect ratio correction
and zoom. So my sphere is really an ellipsoid and rotations are
fast/slow depending on direction. Should be fixable.
My main pain points are:
- rotation over the poles
- smooth movement without glitches and numerical issues
- correctly tracking the anchor point, i.e. making sure it's always
under the mouse cursor
Most examples I could find suffered from some or all of these problems
as well, yet google earth and other applications prove the problem is
perfectly solvable.
If anyone can provide a working sample or some pointers I'll greatly
appreciate it - this problem has really gotten frustrating by now.
The question already came up on this list in 2008 and Rob wanted to
write an example, did that actually turn up anywhere?
Help? Please?
Sören
