From: azuma@cs.unc.edu (Ronald Azuma)
Subject: Re: head tracking
Date: 1 Dec 91 20:25:49 GMT
Organization: University of North Carolina, Chapel Hill



In article <1991Nov27.002110.24093@milton.u.washington.edu> brucec@phoebus.
labs.tek.com (Bruce Cohen) writes:

>Could you explain why you decided to put the LEDs in the room, and the
>cameras on the head?  I would have thought that putting the LEDs on a
>headband or some such would be easier on the user, and would allow
>better coverage.  

	Briefly, the reason is to get better accuracy with the same resolution
sensors.  Imagine trying to detect the change in the positions of the LEDs on
the user's head after the user does a 0.5 degree rotation, when the cameras are
mounted a few meters away.  But if the cameras are on the user's head, that
same rotation generates a large apparent motion of the LEDs on the ceiling,
which is more easily detected.

	To get similar resolution from an outside-in (LEDs-on-the-head,
cameras-on-the-ceiling) approach, one could imagine putting lots of cameras
in the ceiling to compensate, but that would be costly: adding more LEDs is
cheaper than adding more cameras.  Alternatively, you could separate the LEDs
by long distances (making the rotation "lever" larger), but the resulting
"antlers" will be cumbersome and require wider field-of-view lenses on your
sensors, spreading out the resolution.

	Note that if the user is constrained to a small working volume, the
outside-in approach is preferable to our configuration.  We take the inside-out
(cameras-on-the-head, LEDs-on-the-ceiling) approach because we're exploring
scalable tracking systems.

	The references previously listed provide more details.

							Ron Azuma
							(azuma@cs.unc.edu)