UntoldDreams > I try to understand your arguments, but I can't convince myself that you're right.
Not on the conclusion: I agree that Move has a better real-space positionning system in the reference frame of the TV set. I just disagree with the way you see it. But maybe that's just a misunderstaning.
I'm still convinced that both technologies are the same, but with a reference frame change. The fact that the camera moves or not is not the main issue there, I think.
Here's a different attempt to explain myself. Let's assume that we have a camera lookint at an object (be it wiimote IR sensor looking at the bar or the PSEye looking at the sphere)
A] camera gives a near-perfect azimuthal angle and polar angle (theta, phi)
B] if the object is symmetrical, size of the object gives an estimation of the distance of the object (r)
B'] if the object is not symmetrical (sensor bar), camera gives a mix of the distance and on of the angles defining the orientation of the object (yaw at "central position"). It's tricky, but as long as you're close to the "central position", it's second-order in yaw, so you get distance. You'd need a frame instead of a bar to remove this ambiguity.
B''] if the object is not symmetrical (sensor bar), you're also get another angle (roll) of the orientation of the system
C] acceleration further improve the computation of the distance by comparing motion observed and acceleration registered. The fact that the camera move and record acceleration, or the object move and record acceleration is no different at all, it's a simple reference frame change.
D] orientation of the moving item is recorded by gyroscopes / magnetometer / gravity. Let's call those angles a, b and c.
For the two system,
theta and phi are very good, r is also good. a, b, c are less precise and prone to error.
Now, it depends on whar you're wanting to do:
- if you want to get X, Y and Z: Move is better, since X, Y and Z only depend on theta, phi and r. If you want to compute X, Y and Z with wiimote system, you also need a, b and c (which are especially unprecise if you don't have a wii motion plus...) Moreover, if the sensor bar is not visible, you don't have those theta, phi and r => awful.
- if you want X' and Y', the coordinates of the intersection of a line tied to the Move/Wiimote and the screen plane, Wiimote is best, since it uses only theta, phi and r, while Move also need a, b and c
As a conclusion, it's obvious that wiimote is quite bad at 3D positioning, first because keeping the sensor bar in the IR sensor direction is impossible, then because a, b and c (required for X, Y and Z absolute computation) require orientation.
That would be vastly different if you replace the sensor bar with a full 3D frame. I've done motion tracking of an object moving, tied to a camera, that worhs perfectly. But the camera was looking at 3D unplanar objects, not at two points. There's even no need for gyroscope/accelerometers for that.
It's the same for Move, you don't need gyroscopes, accelerometers and others if you replace the single sphere by four spheres arranged in a 3D fashion. This way, you have full 3D positionning with only the camera. Problem is that a full frame is too big to put on a device in your hand.
I hope I'm more clear...
UntoldDreams said:
There is a big difference between using a fixed point camera to calculate displacement and a free form bouncing rotating camera. Its quite complex unless you have a million dollar infinite high resolution camera.
I've done it with a (good) webcam and a $300 camera... No problem, really. But not with two points, you need four, or you need to know the orientation of the camera.