mattski2

I've just been talking to a guy who worked
with the team who developed our beloved
speed cameras and so forth...

_)really(_ cool technology and it can do loads
of stuff you don't expect..

1. It can recognize number plates from very
shallow angles in varying weather conditions
and times.

2. Number plates reflect in two spectrums,
infrared and visual and they can recognize
both. There is your clue on how to defeat
them :-) - i.e. some film that blocks both.
If a number plate is obscured they can grab
a percentage of it and interpolate the rest
based on the make, model and colour of the
car, oh and they also have all the V5 doc info
for reference.

3. They can recognize the make, model and
colour of a vehicle and they are self-learning.
If they see a car they haven't seen before
they pull out attributes about the car that
they may have already seen and can basically
know if a new car comes on the market.

If you change the headlights and the badge of
the car it _might_ confuse it a bit but it will
use other characteristics to identify the car,
i.e. the shape of the grill is really strong.

4. They are accurate upto about 110mph in any
weather upto medium rain. Oh and they can handle
heavy traffic in 3 lanes @ 110mph !

5. The car park ones (like at heathrow) can
recognize the make, model, car, plate, you.
They also know if one person drives a car in
and another person drives it out.

6. They can ask the system to track a particular
number plate and also ask where a given car is,
certainly in London.

7. They made systems that can recognize suspicious
behaviour in a group of people or in a car park.

8. The cameras on the M4 are v.good at recognizing
buses and taxis, _dont_ use the bus lane !!!

v.cool but v.scary!

M

darlodge

Does that mean that a lot of this is done by a person instead of a computer? i.e. searching V5's, looking for characteristics.

Either way we buggered

Darren

jjones

If a number plate is obscured they can grab
a percentage of it and interpolate the rest
based on the make, model and colour of the
car, oh and they also have all the V5 doc info
for reference.


tampered video footage? inadmissable in court?

Clarebabes

I worked for a company which developed ANPR - Automatic Number Plate Recognition. It wasn't nearly as good as this, and I am a bit sceptical of it's claims to be honest.

Our software had to get a really good image of the plate at a certain angle and size etc.... Even then it was not guaranteed - about 75% correct. This was actually considered to be above average!

I think this may be scare tactics?? Hope I'm right!

Clarebabes

This is a grey area because types of digital recording could also be said to have been tampered with. If a "real" frame is captured every 8 frames or so, then the rest are built up from only taking the changing areas, then that could be considered to not be real evidence.

The company I worked for took full frames every time and did not do this.

As for ANPR, the numberplate which was recorded was also backed up with a digital image of the car, so if there was any discussion as to the wrong numberplate being captured, the real one could be seen by the image, if you get what I mean.

MATTeL

When I was at Uni one of my friend's dissertations was about character recognition.

He was able using photos (7 years ago and we didn't have a lot of techology to play with) of different number plates at differing distances (fixed lense camera)with different spacing and of characters.

The software he wrote was able to identify everyone. He used fairly sophisticated AI coding to teach the system the letters, the system was then able to work out characters it hadn't seen before for the characters it already knew.

The one number plate it never cracked was one we put together using Paint (hey it was 7 years ago and graphic programs were not as good as they are now) the script was cursive.

If that was possible back then...

BTW his dissertation and code was in the Uni library as it gained a First!

mattski2

nope, it's not a wind up! he wouldn't tell me a lot of the stuff because it's classified info (it has military applications as well) and yes, it does do all of that - the guy is a brain on a stick

re: the V5 and interpolation thing, it's all automatic, and the systems can go into a learning mode where they learn more characteristics and stuff. It can also flag pictures for further investigation by a person and stuff.

Clarebabes

Was that stationary though? Bit harder when cars are travelling at between 50-100 MPH on a three-lane motorway!

Hope he got a bloody good job.

mattski2

nope, it can handle high volume traffic accurately @ 110mph over 3 lanes in medium rain day or night, of course it will miss plates if they are obscured by other vehicles etc etc but it's pretty impressive stuff, oh, and it can recognize the front and back of vehicles, including bikes...

MATTeL

He had to use photos Clare... I mentioned he took pictures of the plates (almost got into trouble a couple of times).

The speed of the car has nothing to do with the recognition software. It has everything to do with the quality of the camera and the speed of the film.

Yes... he did also take pictures at night with flash to try to obscure the plates, fog and rain were also introduced.

He was and probably still is a very bright bloke and very nice with it. He walked straight into a very good job.

He got a 2:1 as a final result, that piece of work earned a first. He didn't want a 1st for his degree as that would sound like he didn't enjoy himself when he was at Uni!

(edit to change god to fog... using a diety to obscure a number plate is going a little far! )



[Edited by MATTeL - 6/14/2002 12:12:08 PM]

Clarebabes

Mattski, I was asking that of MATTeL's mates software, I know it is possible in commercially available software to recognise moving numberplates.

The company I worked for also was developing facial recognition software which was very hit and miss. I don't see how software could possibly do this in poor light, especially with the amount of different expressions our faces have!

Blimey, this is getting bad! Probably use it to catch all those real criminals like people who go 5 mph over the speed limit etc..... [img]images/smilies/mad.gif[/img]

scoob_babe

been well sickening for the rest of us being a right party animal and still getting such a good result

Clarebabes

I did bu66er all work, was a party animal and still have the debt

Got a 2:2 though with no work at all! I often did all-night revision before a morning exam!!

scoob_babe

oh well, shame I'm the least academic and dimwit of the family only getting a 2:2
but smart enought not to have any debts (now or at the time)

davyboy

smart or boring

scoob_babe

but didn;t have to rely on 2nd hand sale of scoob though

robski

not exactly related but,
approx 15 years ago (!) my computer science lecturer had a contract with the US military as she was a **** hot AI person

she told us of a system that the yanks were trying to develop that could spot hidden tanks from high altitude and fast flyby video footage. (this was when we expected USSR to flood us with tanks dont forget)

In not sure now, but in those days AI had to be "trained" and the start was very key as to how the AI developed.

The yanks spent 1 day filming a wood without tanks in it, and towards the end of the day added tanks with various camo disguises and in positions etc.

Anyway after a few weeks they were struggling to work out why the system was VERY unreliable even with obvious "easy spots". Took em a while to work out that the AI had learnt when tanks were present from the shadows generated in the wood. It was the most obvious change the software could detect, so it had developed AI based on this, then it was easy to spot em, based on direction of shadows

this was early AI, and I should imagine its a advanced a bit by now lol

(oh and the end result was a complete refesh of the AI, and some much more simple starting data was fed to avoid the same learning pattern)

robski

Little Miss WRX

You changed god to fog?

davyboy

oooooooohhhhhhhh bitchy!

Get to the Nurburgring and Donnington SIDC track event, and we can sort this out properly.

Clarebabes

Thought he meant acts of God...

Maybe hurricane, floods and plagues of frogs???




P.s. common sense when opening one's mouth, (or typing) is more important than academic qualifications.

[img]images/smilies/mad.gif[/img]

[Edited to change stupid error - that's what being dim and having debt does to you]

[Edited by Clarebabes - 6/14/2002 12:20:18 PM]

MATTeL

Edited again Michelle... btw get back to your own forums!

carl

robski -- neural networks is what you're talking about. Getting the right training sets (and lots of them) is essential to making it work.

MATTeL

Clare - I mistyped a word and I get slammed! (okay 2 of three letters wrong but still)

BTW you don't know what my qualifications are!

[Edited by MATTeL - 6/14/2002 12:18:34 PM]

carl

Shouldn't it be 'more important than' ?

Sounds like you're not typing much common sense to me.

[Edited by carl - 6/14/2002 12:17:52 PM]

davyboy

Easy girls!

Little Miss WRX

We moderators look out for each other's forums as well as our own - tis called team work

Clarebabes

Ok, ok, I realise that it should have been "than" rather than "to"

MATTeL, I wasn't referring to you!

This started out as an intelligent and reasoned thread. Now it has been lowered to petty levels. [img]images/smilies/mad.gif[/img]

Little Miss WRX

Nor are we interested

scoob_babe

Davyboy, I would rather spend my time with human beings / friends than spend all my time with a lump of metal

MATTeL

Carl - If I remember back to the Knowledge-Based Systems lectures neural networks either phsical or coded are used to produce some sort of AI.

So much trying to work out the cannibal, missionary crossing the river problem using LISP!

My friend enjoyed that and the pattern recognition side of the graphics lectures, hence the software he wrote.

[Edited by MATTeL - 6/14/2002 12:27:36 PM]