Smart Cameras To Predict Crimes

hairybacchus writes: "The Independent News is reporting that scientists at Kingston University in London have developed video processing software that is able to predict behavior patterns of the people on-screen. They say it will be used to alleviate congestion in the London Underground or alert police to potential muggings. I wonder how long it will be before this is combined with face-recognition technology? It's spooky." I can't wait. "We searched you because the computer told us to." Trust the Computer.

Comments

You slashdotters are a bunch of cynics..

[Nevermine]

How cynical can you be.. whenever something like this comes around you predict the end of the world.. It's not a question of somebody getting arrested because they thought of mugging a person on the street.. it's about the ability to do city surveilance more effectively by reporting suspecious behaviour of people on screen.. imagine having to monitor 100 cameras at the same time.. wouldn't it then be somewhat of a relief if the program would sort out the screens that show suspicious events on them? Come on people get real! assuming this camera tecnique would work..

Re:You slashdotters are a bunch of cynics..

[Myco]

I think you're missing something important here. Technology enables ordinary surveillance tasks to be replicated and scaled up by amounts previously unimaginable. The result is not just more of the same -- at some point, it introduces a qualitative change

Consider surveillance cameras on city streets. Sure, the fact that I walk down a particular street at a particular time is public knowledge -- anyone could see me and remember. But what if every step I took in public was recorded on video and tracked? Whoever had that information would know a great deal about my behaviour, and that information could be used against me. Pervasive collection of information, even public information, can be a grave threat to privacy.

Now consider the technology discussed in this article. Phenomena such as racial profiling have taught us that an innocent person can suffer horribly at the hands of law enforcement personnel just because they fit a perceived statistical profile. Imagine a world where everyone is afraid to act in any way unusual for fear of being stopped for "questioning."

And you can forget the argument about "if it works, it's okay." First of all, these methods are inherently statistical, and statistical methods are never 100% accurate. If they were, they would be logical, deductive methods. Statistics is inductive.

Secondly, even if you did claim to have perfect foreknowledge of crimes to be committed, you create a predestination paradox. At what point does a would-be criminal make up his or her mind to commit a crime? Who's to say he or she wouldn't back down at the critical moment, or be unable to go through with it due to some chance event?

My real point here is that we can't always rely upon "more is better" methodology as our technology progresses. We have to consider how scale affects the nature of our technological activities. If we are blind to issues such as these, then eventually we'll get screwed. Maybe this prediction thing will turn out to be benign or even beneficial. But there are many, many issues of this sort, and some of them are going to bite us in the ass if we don't raise hell when we see a problem. Dig?

Re:You slashdotters are a bunch of cynics..

[Jim Norton]

What I am most worried about is the flagrant misuse of these technologies. I came across a very good article where the reporter would spend some time with those monitoring the cameras? What did they do in their spare time, waiting for some criminal to catch? They were watching women walking by, or having sex in their car, or whatever ... how would you feel if your spouse or daughter was being watched by some complete stranger? Would that make you feel comfortable knowing that some sicko could be watching you, your friends, or your loved ones at any given moment? Don't think it can't happen ...

This reminds me...

[mav[LAG]]

of an old Guardian Newspaper ad on TV (a few years back now). It showed a skinhead running towards an old man - then froze.

VO: Some newspapers stop here.

Unfreeze and said Skinhead sweeps man out of the way of falling masonry i.e. it was a rescue and not a mugging.

VO: The Guardian - get the full picture.

I guess with this technology in place, computer-controlled lasers would have taken out the rescuer before he could act :)

Reminds me of a story I heard about...

[WolfWithoutAClause]

A guy I work with has a PhD in image processing. He relates this story of a system that was designed to try to detect human beings, and raise the alarm so that a security guard could check it out; rather than have a security guard staring at it continuously.

Anyway, they wrote some software- it more or less just looked for a human sized blob that moved. Worked too- it could detect human beings pretty well.

Trouble was, they found that it was unreliable- it tended to think birds landing in flocks and groups were people appearing and disappearing. So they improved on the algorithm, and put in some code that if the system could see the wings flapping- it would realise it was birds and ignore it.

Anyway, it worked pretty well, so they thought they'd give a hard test. Could someone deliberately evade it? They got a grad student and told him to work out a way to fool it. They set up the computer guarding a notional prize, and set him at it.

The grad student puzzled over it for a while, then siddled into the middle of view; and removed his jacket. He then waved his jacket over his head vigorously. The computer saw all the flapping, and activated the 'bird' assignment and he was able to steal the item...

I have seen this

[Anonymous Coward]

I have seen this first hand. It's pretty cool. It learns what is "usual" about a scene and then monitors the scene for unusual events. Scenarios include:

Locating "suspect packages" left in public places

Spotting vehicles parked in dodgy places

Watching for people accessing secure areas

Making sure no service vehicles get onto runways

Yes, all this is possible with more conventional technology but these often need a human being in close attendance. This system filters out noise like stray animals, cyclists, etc because it learns what suspect packages, vehicles and aeroplanes look like and also how they move and behave.

and yes... it could be used to spot human behaviours. It appears that someone plotting a crime moves differently to someone just going about their business. This system knows the rules about human shapes and modalities and fluidity of movement.

My view is that the final bit is a bit of spin for the consumption of venture capitalists and is unlikely to be of much use in prime time - so no need to panic yet. It does however raise interesting questions about "reasonable suspicion", evidence and culpability if someone is wrongly detained. Police would no doubt try to shift resonsibility onto the technology, as is their wont.

Similar Systems used for Traffic

[helloRockview]

A company my company recently required a portion of, Nestor, operates Nestor Traffic Systems (NTS).

NTS uses real-time video and neural-network technology at traffic intersections and railroad crossings to predict traffic accidents and enforce traffic violations (bad news for you guys who blow red lights) - kind of similar to the situation in the article...instead of predicting the actions of people, it's predicting the actions of automobiles. There are already many deployments nationwide and lots more being installed.

BTW, the same predictive neural network technology is used to predict all types of financial fraud, including credit card fraud and money laundering.

Re:Better then letting some cops choose on their o

[cascadingstylesheet]

Statistics from observing policemen in some US states and the number of blacks and whites they stop for checks and searches are well known, no point in reiterating them...

Well, the accusations are well known. Then the US Justice Department got New Jersey to "agree" to actually commission a study of the issue, in a consent decree.

The company hired to do the study found that the incidence of speeding varied by race. In a way fairly consistent with the stop ratio.

The Justice department was outraged, has "grave doubts", etc. because that isn't what they wanted to find.

Quick and the Dead

[Mulletproof]

Of course, all this technology assumes that humans don't have the ability to adapt their behavior patterns when performing a crime. The stupid ones will get caught while the smart ones learn what trips the system to "track" suspects and endevour to avoid those actions. True for nearly any aspect of life; from hacking to shooting rockets into space.

But the point with face recognition would truely be a kicker. Once that system acually becomes reliable, anybody with a record notorius enough to have their face mapped would be tracked the moment they entered a store. Assuming you can't obscure your likeness in someway, of course.

Some figures re London surveillance cameras

[Reziac]

There was a TV news magazine article yesterday (might have been on Sunday Morning) about the 2 MILLION surveillance cameras that now infest London, in response to IRA threats. The piece pointed out that NOT ONE terrorist has been stopped by these cameras (but that abuse is rampant). It also mentioned that the average Londoner is caught on camera 300 times a day.

Privacy issues aside, somehow a 0:2,000,000 success:cost ratio strikes me as a wee bit useless, not to mention being an utter waste of tax money and gov't time.

And that doesn't begin to touch the problem of sorting out the mass of data from 300 screencaps per day per citizen.