From: andrew m. boardman <amb@cs.columbia.edu>
Date: Fri, 13 Aug 93 10:12:57 PDT
To: cypherpunks@toad.com
Subject: Spooking of neural nets and image recognition...
In-Reply-To: <199308131408.AB13144@cs.columbia.edu>
Message-ID: <199308131710.AA07902@ground.cs.columbia.edu>
MIME-Version: 1.0
Content-Type: text/plain


With regard to the scope of surveillance cameras, the elevators in the GM
building have cameras pointed down at an off angle; I thought they would
be mostly useless until I was shown the (probably computer-modeled)
contoured shell over the lens; on the far and, it produces a suprisingly
broad and non-"fisheyed" view of anyone facing the front of the car.
Image recognition is currently limited to a book of photos of people they
(security) don't want in the building, though...  When the real cost of
such recognition systems is low enough that such run-of-the-mill
rent-a-cops are looking to buy them, we've really got something to worry
about.  (How nice that Citibank credit card holders are all getting their
faces digitized, for their security of course.)

[cypherpunks content ends about here...]
	    FYI, the George Washington Bridge carries much, probably most,
	    of the traffic into Manhattan and New York City...)

   Most?  Hardly.  Don't forget the bridges and tunnels from Brooklyn and
   Queens, and the two tunnels from New Jersey, and...

Someone I knew in grad school (they, not I) was doing surveys of
Manhattan-NJ traffic, and measured some amazing throughput on the GWB,
compared with the tunnels.  I *was* being incredibly NJ-centric, though...

andrew