[ExI] ai application, was: RE: suv hipsters

[spike]

-----Original Message-----
From: Alan Grimes [mailto:agrimes at speakeasy.net] 
ubject: Re: [ExI] ai application, was: RE: suv hipsters

>> That's step 1. Then we take a photo of an arbitrarily chosen auto at 
>> an arbitrarily chosen angle. Then, I write an algorithm to recognize 
> the wheels, more specifically the rim elipse. We take the ratio of the 
> vertical apparent diameter to the horizontal (minor diameter) to 
> estimate the angle at which the picture was taken. For instance, if 
> the minor diameter is half the major diameter, then we know it is 
> either a 1, 5, 7 or 11 o'clock photo.

;)

>And then you come across a car equipped with 4 wheel steering (4WS). ;)


The algorithms I have invented will not work for every car, only most of
them, and all the ones likely to be used in a crime.  For that application,
the bad guys want the most generic vehicle on the road, which is probably a
white SUV.  There are so many of those on the road no one pays attention to
them.  Ironically, I myself own a white SUV, as does my next door neighbor
and the guy across the street, all different brands, all with a similar
look.  I wouldn't be surprised to learn a white SUV is the ride of choice
for daytime burglars.  This whole adventure started because the local
constabulary sent out a notice that a crime was committed by two lads in a
white SUV.  This they knew because they had a photograph of it, which they
described as a white SUV.  This narrows it down to perhaps 1 in 30 vehicles
on the road.  Knowing the species got it to one in 2000.

However... the local police do not want to find this vehicle.  Think about
it: cities all over California are realizing they must lay off workers,
since they have falling revenues as far as the eye can see and increasing
costs from retirement pensions.  Bad guys on the loose increase demand for
their services.  So now, it is up to the citizen to find the perps and
report it, not to those who are compelled to do nothing, but rather to those
who have a motive to do something: the victims' insurance company.

Thus my notion that this kind of software needs to be in the hands of the
citizens. 

I thought of an algorithm that doesn't require storing all those photos.  We
JPG-ize the photo of the perps ride.  Then we have it identify the rims,
which come in only a few standard sizes.  Four sizes take in 99 percent of
all cars: if the protestors on Wall Street had cars, they would all be one
of these four sizes.  So if we take the ratio of the apparent vertical
diameters of the front wheel to the rear, we can by parallax estimate the
distance of the car.  Alternately, using a second view taken simultaneously,
we can use that parallax to calculate the distance of the car and use that
information to back out which size rim is on the car, of the four choices.
Now we have an absolute metric, the easiest one to get from a photo, since
tires are black, which creates a high contrast easily identified known round
transition at least twice in every side view.

Then, with that information and the ratio of the major diameter to the minor
diameter of the rim, to pick off an angle, so we can than back out the
foreshortening effect.  Then we create a straight line tangent to the tops
of the two rims, and pick off four pieces of information: the width of the
open fenders (2) and the distance to the rear along that line and the
distance to the front opposite corner along that line.

I think with that information, I would be able to catalog a unique signature
for every common car on the road, without having to store all the photos.  I
also think I could write the algorithm to do it.

Tara Maya's husband does this kind of thing professionally.  Perhaps they
use an algorithm similar to the one described above.  It isn't intelligent,
but it is artificial.

spike