Tampa using cameras to scan for wanted faces--

[Eugene Leitl]

On Thu, 5 Jul 2001, Tim May wrote:

> One of the interesting things is that _ear shape_ is one of the best
> correlation features.

There are a billion: skin pigmentation as seen in NIR illumination (my, do
you look spotty), NIR laser scanning of body features (MEMS mirror
galvanometers), including millimeter wave which penetrates clothing (in
development still), voice fingerprint, person-specific word patterns
(Echelon is surely using these on targeted emails), gait and mannerisms
(not even in development, but sure to arrive some day).

The only way to avoid radiating a fingerprint is to use anonymized
teleoperated hardware as meatspace proxy. And of course you can outlaw
these, unless teleoperated robotics becomes very common in the next few
decades (possible, but I'm not counting on it).

> Of course, to measure ear shape the camera has to have a good view,
> unobscured and at close enough range to get a decent number of
> pixels.  (This makes sense, that ear shape would be a good metric.

Sure there are limitations to the current state of technology. The
biometrics are of lousy quality, take seconds to compute on a ~GHz CPU,
and are not generated in an embedded device.

Nevertheless, imaging technology makes good progress with embedding DSP
cores and using hybrid architectures based on silicon retina technologies
as pioneered by Mead.  Because this is machine vision used on moving
objects, it can tolerate dead pixels, allowing you to boost resolution
(Information in a 640x480 30 fps is sure limited, but with CMOS tech like
http://www.foveon.net/tech_f16.html and tolerance of ~5% dead pixels
multimegapixel sensors plus active optics for tracking and feature
extraction with parallel DSP cores integrated into the sensor you capture
a lot of info, and process it in situ, too). As soon as the devices become
sufficiently cheap you can integrate them into virtually anything
(installation costs typically dwarf hardware costs), including street
signs (OCR to read license plates is almost mature), lanterns, copers'
wearables, etc.

The extracted biometric alone is tiny, and can be readily transmitted
using even current paltry 9.6 kBps wireless modems.

> I've been noticing the variations in ear shapes since I heard about
> this scheme. Also, I can imagine the various conformal
> transformations--different angles of view, for example--preserve
> certain relationships well.)
>
> I can believe some kind of automated face recognition is being done
> with points of entry, such as international airport arrival points,
> but I find it hard to swallow that "crowd shots" from overhead
> cameras can do anything meaningful.

With current tech the error rate is still high, but it's for real. The
hardware is going to become better due to Moore alone, including silicon
retina dedicated hardware, the economies of scale will apply, and of
course the software will get better, so the capabilities will be ramping
up very rapidly over the next few years.

> The Tampa action may be mostly social engineering: "We're watching you!"

Wehret den Anfaengen. The capabilities are still mostly vapour, and the
coverage still spotty, but exponential processes have their
counterintuitive dynamics. Twenty years more of those, and you'll be very,
very surprised.

-- Eugen* Leitl <a href="http://www.lrz.de/~ui22204/">leitl</a>
______________________________________________________________
ICBMTO  : N48 10'07'' E011 33'53'' http://www.lrz.de/~ui22204
57F9CFD3: ED90 0433 EB74 E4A9 537F CFF5 86E7 629B 57F9 CFD3