h a l f b a k e r y
Get half a life.
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
or get an account
When videoconferencing two groups,
there often isn't enough bandwidth or
equipment to show
all participants all of the time.
Imagine a device placed in the middle of the
conferencing tables in both locations. It has one
"head" end which it turns towards the loudest
sound source in the room.
camera is placed in its middle pointing to the
head end. It will now turn to the speaker.
This works better for a camera with autofocus,
and there are some details about the algorithm
to get right: the camera should be "lazy"(don't
pan around unnecessarily) yet reactive, and
it shouldn't get tangled up in cable if people
speak in the same order around the table.
The device also needs to be quiet enough not
to interfere with a microphone placed on it.
Maybe this could be done using a stationary camera
and merely a swiveling mirror.
Probably what Dan is talking about [rmutt, Aug 31 1999]
Panoramic video from multiple cameras
They're working on the microphones, I'm told. [rmutt, Aug 31 1999]
Just about exactly what [johan] was speculating on. Note the array mics in the lower image; they turn on the camera showing the person who is speaking. [bristolz, May 20 2002]
||I've seen this done with a special wireless microphone. The camera swivels to point at the microphone. Mostly, this was intended to follow a pacing lecturer.
||We've actually got one of these
in the research building where
I work. It's got four microphones
(phased-array, to locate the
sound) and a Sony controllable-swivel
||The only problem with it is that
people tend to talk back and forth
for several seconds at a time,
so the camera's always swivelling
back and forth.
||One solution would be to have a cluster of cameras in the center of the
room (or table), connected to some CPU that monitors the phased array of
mic's. So instead of lots of camera panning, each camera would cover one
chunk of the room. (4 cameras == 90 degrees) Seems like some of the
panning could also be avoided by zooming in and out.
|| The next cool thing would be to write software for the CPU which would
speculate who was going to speak next. There's probably a
"self-fulfilling prophecy" trap here, though. As the cameras panned
towards the next most likely speaker, the speaker, seeing the camera aim
at it, would then feel compelled to talk. Pretty entertaining, come to
think of it.
|| I wonder how much long-term analysis of individuals voice and speaking
patterns could be done to further "predict" how long someone was going to
speak. The CPU could store the profiles. (E.g. name=johan
||Just strap a camera to somebody's head and voila. no techno-gadget needed.
Aug 22 2001, last modified Aug 23 2001|| |
||I posted the same idea yesterday and jutta linked to this site. I deleted my post -- thank you for notifying me jutta. I would like to suggest that a parabolic microphone, and recent advances in acoustic triangulation and face recognition technology be adapted to enhance this already sound (no pun intended) idea.
||And [Normzone] yesterday as well - however theirs had the distinct advantage of the joke-flower-squirting-nozzle functionality.
||What happens when an unintended sound is made? In a fast panning system, a falling book would cause it to jolt to that direction, confusing people on the other end. In a slow panning system, it would waste time panning towards the book and then back.