Hi all, long term lurker hereby de-lurking....
Dan Fabulich <daniel.fabulich@yale.edu> wrote:
>*How do you know which screen to look at?* There are hundreds of
>thousands, if not millions of cameras you could be monitoring, any of
>which may have a picture of somebody slitting a child's throat. The only
>way to tell which one is to LOOK at all of them, requiring a gargantuan
>amount of processing power/time. That's where the computation comes in,
>and a hell of a lot of it.
But given the technology required for universal surveillance, it should be
comparatively easy for everyone to have a gizmo strapped to their wrist
monitoring vital functions which would trigger an alarm (and nearby cameras)
in the event of their demise. Any murderer would be caught "on tape" and
police would converge on the location immediately. Even if some kind of slow
poison was used, you could simply rewind the video from the victim's own
personal camera, find the point when their food was tampered with and again,
nail the murderer.
Also, I read recently (sorry, no idea where) about software that was being
developed (and already functional) to monitor video-surveillance in real
time. The software triggered an alarm when it saw "suspicious" behaviour. I
think it was still fairly primitive but I seem to remember it was already
capable of recognising the difference between car theft and a normal driver
getting into their car. In fact, it could even predict (presumably with a
fair few false positives) that some one was likely to commit a crime based
on their movements / stance etc.
Dan Fabulich <daniel.fabulich@yale.edu> also wrote:
> But identifying text is a piece of cake compared to identifying people.
> Trying to explain what a person looks like to a computer is, as Eliezer
> might say, just one step above trying to explain it to a rock. Today, we
> don't have anything which can do this anywhere near as well as a human can
> (to the best of my knowledge; this project is hard, but not impossible).
> If you'd done very much programming you'd see why this is so hard. All
> the computer knows is a grid of numbers (a grid which, as we happen to
> know, corresponds to a grid of colors). Based on that alone, the computer
> has to identify what a person looks like. Notice also that the computer
> can only use simple arithmetic and set theory to process those numbers. So
> I want you to imagine a function which uses only arithmetic/logical
> operators, (+, *, =, OR, NOT) and membership relations, (X is a member of
> set Y), takes as its input a grid of numbers, and yields as its output a 1
> if the grid of numbers corresponds to a picture of a person, and a 0
> otherwise. Can you see how hard this is? It's not impossible, obviously,
> but it's Very difficult. That's why we still haven't done it thus far.
Actually there's a system on trial in some nightclubs in (I think) Holland
which automatically compares photos of people who have been banned with
incoming patrons and sounds an alarm if it recognises them. Again, this is
based on a hazy recollection of a news article but I seem to recall that the
system was even quite good at seeing through disguises.
Anyway, the point is you can already (to some degree) automatically turn
video into semantic information and as computing power goes through the roof
(as I think most on this list would agree it soon will) this will become
more and more easy. At the moment speech-to-text software is also rapidly
improving (though still prone to error) but it's already got to the stage
where I'd probably dictate a book length piece to my computer rather than
type it.
Oh well, just my two cents...
Mike Steven
This archive was generated by hypermail 2b29 : Thu Jul 27 2000 - 14:06:48 MDT