Note: Where available, the PDF/Word icon below is provided to view the complete and fully formatted document
Disclaimer: The Parliamentary Library does not warrant or accept liability for the accuracy or usefulness of the transcripts. These are copied directly from the broadcaster's website.
Facial Recognition -

View in ParlViewView other Segments

NARRATION

During the riots in England last year, CCTV caught criminals in the act, twenty-four-seven.
Scotland Yard released hundreds of images to the public to try to identify the rioters. Despite
tens of thousands of man-hours chasing leads, they solved just one crime per thousand cameras. Now,
new facial recognition technology promises crime fighters their greatest gift.

Prof Brian Lovell

This is a huge application because we can identify someone within milliseconds.

Anja Taylor

Just about everything we do these days, we're being watched. There are cameras everywhere. But just
how difficult is it to take a grainy, low-quality image of someone and find out who it is?

NARRATION

Trawling through surveillance footage takes hours of police time. Pinning down a suspect takes even
longer.

Senior Sergeant Kevin Darch

The quality of the images are very poor, by and large. To try and find enough detail in that image
to make a reasonable comparison is a very time-consuming process.

With this kind of image, we have made a comparison between this one and a lot of others. There's
been a lot of unlawful transactions made over a period of time, and so we have a suspicion that one
person has been involved in a number of different crimes.

Anja Taylor

And how long can it take you to identify a person from footage like this?

Senior Sergeant Kevin Darch

Sometimes many weeks of comparison.

NARRATION

Researchers at the University of Queensland believe they've cracked the problem.

Prof Brian Lovell

So here we have two thousand mug shots.

Anja Taylor

Right.

Prof Brian Lovell

Now pick a face from here.

Anja Taylor

Okay, how about that lady there?

Prof Brian Lovell

Okay, now here's four thousand identities. They're not the same photos, but they're all in here.

Anja Taylor

Right.

Prof Brian Lovell

I want you to find that person in here.

Anja Taylor

Right, okay. Is it the same photo?

Prof Brian Lovell

No it's not the same photo. They could be years apart.

Anja Taylor

Ah ... I give up.

Prof Brian Lovell

Give up. Well that's human face recognition. So we do this in milliseconds.

NARRATION

Professor Brian Lovell and his team have come up with a facial recognition system that can identify
someone from low quality vision.

Anja Taylor

'Unknown'. Okay, so it doesn't know me.

Prof Brian Lovell

That's easily fixed, so I'll just add you now.

Anja Taylor

Okay.

NARRATION

Brian captures a low-res image of me. Now I'm in the database.

Anja Taylor

So it's recognising me. Wow, that's amazing.

Prof Brian Lovell

That's right, no it doesn't take long.

Anja Taylor

So even if I turn around that way or that way, it still knows who I am.

NARRATION

The software has to find a match by sifting through a bank of faces. These could be persons of
interest, like a mug shot database.

Prof Brian Lovell

It's like a web search engine, so it can search through very large numbers, and it's almost
instantaneous.

NARRATION

It's impressive that I can be recognised standing still, but even when I'm walking and looking away
from camera, it still knows who I am. Not only that, it's even tracking me from camera to camera.

Prof Brian Lovell

Non-cooperative surveillance is the holy grail of surveillance. You don't want people to know that
they're being photographed. So this is a very major breakthrough.

Anja Taylor

The system borrows from a language processing technique called 'bag of words'. Just as a sentence
can be viewed as an unordered collection of words, a face can be viewed as a collection of
two-dimensional tiles.

Prof Brian Lovell

And then it can very rapidly compare the templates from the faces, so it can search through you
know, a hundred-thousand person databases. The inner measures are concentrating on the skin around
the face, the outer measures look at the ears, the earrings and these sorts of things, and they
have different weightings. Hair is not such a problem, unless it's covering the face. Normal
spectacles aren't a problem. Sunglasses, we can manage to some extent.

NARRATION

Even when I'm in a group, it can still single me out. And it's not just recognising me, it's
identifying all of us.

Prof Brian Lovell

We can process about twenty people simultaneously and identify all of those. We have done trials in
airports with like four-thousand passengers, and they've been very successful.

NARRATION

Before this system, the ability to track individuals from camera to camera didn't exist.

Prof Brian Lovell

We can typically get a ninety-five per cent cross-camera matching between two cameras. So the
chance of somebody slipping through a whole set of cameras is very low. The obvious application is
in terms of like airports and so on, for tracking passengers or looking for particular passengers.
And for measuring passenger transit times and so on.

NARRATION

The technology can also be used to analyse recorded video from CCTV, such as footage from the
England riots.

Senior Sergeant Kevin Darch

Facial recognition technology will be a fantastic tool for us to solve a lot of crimes.

NARRATION

And with real-time recognition, there's the potential to prevent crimes from happening in the first
place. But with the nature of this technology being non-cooperative, could facial recognition
technology be a privacy disaster waiting to happen?

John McAteer

One of the things that's being proposed is a facial recognition system that could be used across
the whole internet. Now that offers a very, very vast pool of data which can be used, perhaps to
track somebody. And that's a real concern. And that really gets back to one of the fundamental
notions of privacy, is that there's a notion of liberty and the right to be left alone.

NARRATION

Another concern is the huge archive of footage that can be retrospectively analysed.

John McAteer

When you walked down the high street ten years ago and there were CCTV cameras, you did not consent
to it being on a very large database. I think that broadly offends the public interest.

Prof Brian Lovell

Like any technology, it depends how you use it. So once again the technology has to be well
controlled.

NARRATION

Like it or not, facial recognition technology is currently being rolled out. And with more and more
sophisticated surveillance, we may have already surrendered our privacy in public places.