Slashdot Mirror
Video Surveillance Tech Detects Abnormal Activity
Repton writes with news of a company, Behavioral Recognition Systems, that has received 16 patents on a new video surveillance application that can convert video images into machine-readable language, and then analyze them for anomalies that suggest suspicious behavior in the camera's field of view. The software can 'recognize' up to 300 objects and establish a baseline of activity. It should go on sale in September. "...the BRS Labs technology will likely create a fair number of false positives, [the CEO] concedes. 'We think a three-to-one ratio of alerts to actual events is what the market will accept,' he says. 'We could be wrong.'"
Great! Now, all they have to do is combine that with this, and we can all sleep soundly.
Caveat Utilitor
If I walk past a security camera in a full sized squirrel outfit, humans couldn't even figure that one out let alone a computer. These systems are just dumb. Wait until computers are smart enough to talk with us to develop something like this otherwise it'll never be remotely accurate.
Google's Super Secret Search Algorithm: SELECT @search_results FROM internet WHERE @search_results = 'good'
As long as the cops don't beat too many people too extremely for false positive behavior I can't see where this could be a problem. And Homeland Security is already working on getting some Executive Orders written up that will make it a crime to act in ways that cause false positives, so there should be no false positives in the near future (by definition they will be real positives). Problem solved.
I'm an American. I love this country and the freedoms that we used to have.
I hope they don't bet the company on these patents since, as was discussed on ./ less than 24 hours ago, software patents involving digital signals may be invalid. (The field is generally called Digital Signal Processing.)
While it's a worry how 1984 society is becoming, I don't think false positives are a particularly bad issue with this technology. An operator who would normally be supposed to watch multiple streams of video for anomalous activity can use these more like bookmarks for subsequent human verification. The bigger issue as I see it is that 16 new patents were just granted on software/algorithms.
Me lost me cookie at the disco.
I thought what I'd do was, I'd pretend I was one of those deaf-mutes.
... when it first gets deployed - if it gets deployed - spread the word across the internet and get people to regularly silly walk past it and do other wierd but non threatening stuff. Hey presto, so many false positives it's rendered useless.
And you don't think that will soon be made illegal? You sure sound like a terrorist to me, to Gitmo with you!
I'm an American. I love this country and the freedoms that we used to have.
Many of you will concede that video surveillance is desirable sometimes. And that security is desirable sometimes.
This is a tool that can help surveillance. Perhaps wrongful, privacy-violating surveillance. But very possibly surveillance that we want.
I'd say a 1/4 error rate is fine - it still narrows down the amount a human would have to screen, and that kind of error rate will pretty much insure a human is in the loop.
1. File a bunch of mumbo jumbo patents on video surveillance
2. Convince some trade rag to interview your CEO
3. Submit the story to /. as a clear and present danger to "Your Rights Online"
4. ?
5. Go public!
I swear you must be a troll from Homeland Security...
That's like saying "Oh sure, it is worrisome that I have a live hand grenade with the pin pulled jammed in my mouth, but I don't think it would be extremely bad if it just blew off one of my pinky toes"
This kind of technology makes me want riot...ahem...i mean...to exercise my 1st amendment right to protest in a law abiding way.
I'm sickened. The CEO says: "We think a three-to-one ratio of alerts to actual events is what the market will accept."
Thank you Dave Raggett
I just was just viewing this right before logging onto slashdot and seeing this topic.
I'm sure everyone on Slashdot is donning their tinfoil hats and screaming big brother (I've already seen a couple posts to that effect) but that really isn't the target market. You'll find that by far the most customers of CCTV equipment are private companies. Pretty much any large store will have an extensive CCTV system to watch for shoplifting.
Ok well the problem is that you have to have humans watching it for suspicious activity. It is completely infeasible to hire one human per camera, and the more cameras a given human has to watch, the less they catch. Well, something like this could help. If it sees something suspicious, it brings it up on a display to one of the security personnel. The person then decides if it is a problem, or a false alarm.
A moderate amount of false alarms is fine. This wouldn't be a case of "The system went off, arrest him!" It'd be a case of "The system went off, let's have a human watch and see what's going on." It would allow for better use of security personnel.
Heck, I'd be interested in a system like this at work. We have CCTV on our computer labs. However we don't have anyone monitoring it. It's more for liability reasons, and so that if someone steals or damages a computer, we can hopefully help the police catch them. However prevention is better than clean up. So it'd be cool if when the system thought something was wrong, it'd notify staff and we could look. If everything was fine, we carry on as normal. If something is indeed happening, we call the police.
You've got to stop with the idea that these sort of things are designed to figure out what you are thinking for some evil government plan. They aren't. They are designed to help make security systems more effective.
As noted in TFA and if the false positive ratio can be reduced to even 10-to-1, this technology might rapidly become the best friend of the fellow who has to constantly scan 100 surveillance screens for unusual activity.
But this system's definition of "unusual activity" intrigues me. If one of these toys is set up for example in a bank to monitor a vault door and a bank guard passes by the door every hour on his rounds, the software would presumably record that as "normal" activity. What is the "unusual" element that would prod the AI into sending an alert if a thief did exactly the same thing? What dynamic does the system employ to determine if a bank guard is a legitimate bank guard or Willie Sutton? The time it happened? Facial recognition? The fact that the "bank guard" pulled a cutting torch or dynamite out of his backpack and started going to town on the vault door?
Could the system be configured to send an alert when an expected activity didn't occur such as a bank guard or jailer missing one of his rounds?
Its difficult to imagine this system being used in any sort of serious anti-shoplifting capability in a retail setting. Would the AI be able to tell the difference between a customer picking up an iPod to look at the fine print on the box and a shoplifter shoving the iPod under his shirt? Would the system alert on me if I innocently tucked in my shirttail while walking down an aisle? As always, the devil is in the details.
No mod points, no meta-moderating/Firehose/all the other free work Slashdot wants me to do.
Does it detect ghosts?
Mit der Dummheit kämpfen Götter selbst vergebens
UK citizens are safe in the knowledge that they're being watched. That's it citizen, keep moving. Nothing to see here.
Ignoring the Orwellian references, could this be used for some good?
ilovegeorgebush
Abnormal activity? You mean like a slashdotter outside, in the sun, with a date?
Table-ized A.I.
Sure, the market might accept a three-to-one failure rate. But what about the 3 guys who get locked up in a DHS holding cell, get waterboarded and then released without any means of setting the record straight just because they were lost in a shopping mall with a cakebox in their hands?
Machine: Uh oh, Jeremy stoped sticking his finger into his nose. I better call this one in!
...
Operator: Ummm... Why has it shown me Jeremy just sitting there?
Machine: Nope, there he goes again, digging away at his nose, everything back to normal. Better stop transmitting.
Machine: Whoa, he stopping barking for boogers again! Better show the boss!!
Operator: Why does this dumbass machine keep showing me Jeremy just sitting there for goodness sake...
Machine: Boss! Boss! Come on, look! DIFFERENT! ABNORMAL!
Operator: *Hmmm what's for lunch...*
Moved to http://soylentnews.org/. You are invited to join us too!
Anyone else read that as "Video Surveillance TechNICIAN Detects Abnormal Activity"? I was confused for a bit.
Imagine deploying this system at a furry convention.
"Liberty may be endangered by the abuses of liberty as well as the abuses of power." -- James Madison
'We think a three-to-one ratio of alerts to actual events is what the market will accept,' he says. 'We could be wrong.'
So, how many genuine terrorists have we caught in the last few years? Our current rate is several million to one. Three to one sounds good.
If that's what they're shooting for, then I have a name for the system:
Cry Wolf!
Because, that's all it's really going to do!
Heaven help any street performer that gets caught by this video frankenstein's monster, because the cops will, in some jurisdictions, come in blasting away and a mime is a terrible thing to waste!
Apparently the answer is:
"unusualisnotabnormal"
There, fixed that for you.
"I have downloaded hundreds and hundreds of records, why would I care if somebody downloads ours?" Robin Pecknold
I call shenanigans on this.
There is no way they can recognize 300 objects in real world conditions. I work in machine learning (academics) and the current record for generic object recognition sits at around 54-57% for the Caltech 101 database (contains images of 101 different objects). So basically the algorithms of the best and brightest minds in academia (LeCun, Poggio, Lowe, etc) get it wrong half the time !!
If any government officials are listening... Please don't waste our tax money on this !
A million monkeys and this is the best sig they could come up with...
IF color = black THEN
watch(person)
END IF
Consider the 1970s era jet fighter Phantom II F-4. The leading edge of the wing required an extender to help it pull out of dives. The outboard wingends required an upward angle to stop flat spins. During development, the fuselage design was expanded to accommodate the fuel tanks the beast would need, and provision was made for up to three under-wing external fuel tanks. The nose had to be extended to make room for the combat radar defined in the original requirements.
And it had to fly with 4 degree nose-up pitch to maintain level flight.
When I worked on this ugly bird in the 1980s, we said it was actual proof that if you put engines big enough on anything, you can make it fly.
In all likelihood, that's what's gonna happen with all the gov't facial and behavioral "recognition" systems that are under development. They'll pour billions into development, and wait for the computing horsepower to catch up and make it possible to achieve "3-1 false positives" (25% ???). It'll be a monster, but they'll be able to claim "mission accomplished".
Pavlov wouldn't be so famous if he'd used a can opener instead of a bell.
Great so now we'll all have to watch our basic movements and how close together we have certain objects. The future of freedom is looking so wonderful. Sigh....
What application DOESN'T store video in a machine-readable language?
To the fool who tagged this 'unusualisnotabnormal':
You're wrong. Unusual and abnormal mean essentially the same thing - something out of the ordinary, something not routine.
If the point you were trying to make is that authorities shouldn't be suspicious of every unusual occurrence, then perhaps something like 'unusualisnotwrong' would have better served your purpose.
From TFA: $1,500 to $4,500 per camera.
100 screens, assume each one rotates thru, um, 5 cameras, that's 500 cameras, say $3000 per, 1.5 million bucks. You could actually HIRE HUMANS, say 20 of'em, at $75K each to watch 5 screens each, and have a 1:1 ratio of accuracy.
Pavlov wouldn't be so famous if he'd used a can opener instead of a bell.
The problem with this type of software is that it relies on a human being to provide the initial definition of what is or what is not abnormal behaviour.
A human being then points the camera and evaluates the results.
When we can put the cameras into the offices of the police and security forces, and into the politician's offices, and if we can then define which behaviour it is that we don't like in their activities - an inclination to use lethal force on demonstrators; an unearthly preference for military solutions over negotiation and compromise; a predilection for control over others etc. - then I think we will have made a positive step with this type of technology.
When all that is happening is that those in authority are making it easier for themselves to impose their own will on the rest of us, I have no faith in it's use.
I walk with a limp. Does that constitute abnormal behavior according to this software?
just like the DVD became a nanny for kids, the tazer is becoming the defacto response for law enforcement. just like the airbag replaces the seatbelt for a lot of adults, the telephone has slowly turned into a tracking device for parents. (http://blogs.computerworld.com/node/2744) abnormal activity detection technology is likely to become a surrogate for actual law enforcement and security. if "motion zones" in cameras provide multiple false positives, which they do, this technology will find itself likely ignored just as quickly by an operator.
Good people go to bed earlier.
...once it becomes known what sets it off.
Warning: this article may contain humor, sarcasm, parody, and perhaps even irony. Read at your own risk.
Can each observed object in the machine code feed be represented by a stream of shifting symbols streaming vertically down the display? Can you make them green?
I'm tempted to send off a resume. If they're hiring people stupid enough to claim that video is difficult to make machine-readable, as president, I've gotta be able to get a good position there.
Of course, it is impressive to offload that much analog data to a machine, and process it in real time (I wonder what kind of hardware this needs) but the wording of that statement is so fundamentally flawed that man should not be allowed to speak in public (and possibly not claim to be any sort of technical expert.)
> You could actually HIRE HUMANS, say 20 of'em, at $75K each to watch 5 screens each, and
> have a 1:1 ratio of accuracy.
Never pulled guard duty, did you? you won't get a "1:1 ratio of accuracy" even if you hire someone for each individual camera.
Warning: this article may contain humor, sarcasm, parody, and perhaps even irony. Read at your own risk.
I wonder if he got his idea from the Matrix, when Neo realised he was the one, and came back from the dead.... everything looks....green!
A three-to-one ratio of false positives is lower than most paranoids will accept.
Granted we have most likely never met, so I am having to make a sensible assumption about your physique. I suppose it's also possible that I know less about squirrels than I think I do, but I bet there's no way you can get into an outfit the size of a full-sized squirrel.
...wide area surveillance system (which included video analytics for loitering, wrong way traffic in a crowd, crowd panic analysis, smoke/fire detection by video, et cetera) I can point out that there are MANY companies, corporate research groups, and universities that have been doing things like this for several years (in some cases almost 10 years.)
This company is in for a rude awakening when they realize that (a)their price per camera is extraordinarily high (this one metric is the biggest decider in large installation proposals [along with whether or not you have to mount the cameras on poles or just hallways/buildings]) and (b)a false alert rate of 3 to 1 is TOTALLY unacceptable. The entire purpose of video analytics in a security environment is to reduce the workload on the monitoring staff (and hopefully put more of them out into the field) while being able to scale up your coverage. I assure you that a 3 to 1 false alert rate will result in zero customers in a year. Measuring the false alert rate is also highly subjective. Companies tend to use a given scenario repeatedly to measure their results when, of course, this has little to no bearing on reality. Things like the weather (moving shadows affect certain algorithms even when accounted for algorithmically, headlights, flashlights, camera flashes change things, wind, rain, snow, bugs, everything you can imagine, lol...) negate all of these measurements.
It is nice to see new blood in this space, but I hope they were smart enough to make their software offerings totally distinct from their hardware (many companies do not) so that they can integrate with other systems without to much work. That's the best way to make money in the video analytics market right now. The big boys (like SIEMENS) got into the game about 3 years ago and they'll squeeze you out every time unless you can offer something that helps them land a big deal.
Loading...
Yes, this is why any such technology must be viewed with some concern. In parallel with requiring people to change their behaviour for the benefit of the machine, we also have the danger of trusting the machine. Sooner or later, some jobsworth will decide that a beep on the machine constitutes "reasonable grounds" for suspicion, which is all that is required to stop/search or arrest someone in some places. Ironically, a 1-in-4 failure rate is probably a good thing here, since at least then such a decision is likely to be overruled by a higher authority sooner rather than later. But what is ever an acceptable failure rate, given the negative consequences for the innocent victim of the system?
You need a pretty severe penalty/compensation/appeals system to overcome the downward spiral once you start to trust the machine, perhaps something along the lines of exponentially increasing compensation payments with each false accusation of the same person and personal responsibility on the part of the operators so they could go to court and be subject to sanctions if they abuse their position. But of course, this sort of thing doesn't really happen. What really happens is that the Powers That Be, whether government or corporate interests, pretend it's OK for a minority of cases to be wrong, and most of the little people don't have the resources to fight the abuse.
Meanwhile, the person on the other end gets to be like the lady in the UK who was on a TV report recently because her car number plate was cloned: she is receiving automated fines for motoring offences at a silly rate from various government agencies, each of those fines has to be challenged individually in court, and there is no mechanism available to flag as suspect the record in the DVLA database those agencies used to find her. Her life has become an ongoing, government-sanctioned harassment campaign, and while there ought to be one hell of a due process lawsuit in the works since this is the government doing the dirty work, at the time they showed the report no lawyer had been able to advise her on how to stop the madness.
Regardless of the theory, for every mass surveillance technology we allow to be introduced, stories like that one are easy to find. And most mass surveillance technologies don't really work very well anyway: city centre CCTV hasn't been proved to reduce crime, ANPR cameras result in the real criminals investing in false number plates, etc. We would do well to remember the old saying that man should not be judged by machine, and to oppose any sort of machine-based summary "justice", lest we slide further into the hole we are digging for ourselves.
If you disagree, post your argument. (-1, Overrated) isn't your personal censorship tool for views you don't like.
Perhaps I am misunderstanding the situation, but a proposed upcoming "war of machines" seems a bit premature. The real issue in machine warfare, the one which really needs to addressed, is that machines must be able to repair machines. As it stands, each of these devices will need routine maintenance and, as such, are entirely beholden on their creators. Currently, a robot cannot diagnose the strange rattling in a car engine or why the darn "check-engine" light won't go off (and yes, there are computers that help with that, but they don't really fix things, do they?).
No, I think the real concern is not automated "booby traps," I think the real concern is when I drive to a mechanic's shop and am greeted by the ED-209 refitted to change tires. Granted, my lube job would be faster and cheaper, but I'd miss that service with a smile.
http://www.allen-poole.com/
isee what you did there
Notice that the spokesperson says "We think a three-to-one ratio of alerts to actual events is what the market will accept."
What he doesn't say is what their actual false positive rate in testing was... or any reason for believing the false positive rate will, in fact, be anywhere near that low.
I've come to be very leery of that sort of hypothetical statement... ones that lead you to think something has been said that hasn't been said. He's saying that 3:1 is a plausible goal, but he's not saying they've achieved it!
And what does he mean by a "false positives?" Let's say the system is installed to detect (say) humans trespassing on private property in a wooded area. Let's say that no deer had wandered through during the period when the baseline was being established. One day a deer wanders through and an alarm is triggered. Would BRS count this as a false positive, because it was not a human intruder? Or would they count is as correct functioning, because it was "unusual activity?"
Much of what is said in the article is pretty hard to believe on the face of it, at least not without further explanation.
"The BRS Labs software can establish a baseline in anywhere from 30 minutes to several hours, depending on how much activity the camera recognizes and how regular the patterns of behavior are." But it talks specifically about "wooded areas," i.e. outdoor scenes. Will it be able to ignore snow, branches falling in a windstorm, etc. if those events did not occur during the training period?
On that "busy highway," where the baseline can be established in "half an hour," will the software know that the bicycle race or Fourth of July parade is not unusual?
"How to Do Nothing," kids activities, back in print!
http://www.cyberdynegauges.net/
ALL YOUR GAUGE ARE BELONG TO US
(caps filter workaround)
"When information is power, privacy is freedom" - Jah-Wren Ryel
You mean like the humans in the loop with the "terrorist watch list", the ones that tag little children and congressman and ww2 vets and so on for special security checks? The list with the "humans in the loop" that by some accounts say that there are a million terrorists floating around the US? Hundreds of thousands to a million and counting, they won't say for sure,. but those humans in the loop are..loopy, completly and utterly incompetent at best, dangerously fascistic at worst.
If there were even *one hundred* actual terrorists in the US, really just one hundred active dedicated terrorists, we'd be hearing about attacks all the time, let alone a thousand or two thousand or hundreds of thousands. Do people really believe there are that many "suspicious terrorists" travelling around? It's complete and utter bullshit lies and any defense of it falls into the seriously fascist and pushed propaganda realms.
They can't even get people removed who are obviously not terrorists right now, so where is any credible evidence at all that just adding more bogus "suspicious activity" from video cameras and computer algorithms everywhere is going to improve the situation? They are working on facial recognition, putting names to faces, now if you look "suspicious" on some camera feed, then get facially recognized, you could automatically go on some "potential enemies of the state" list, which is what the terrorist watch list really is? This stuff is designed to threaten, cajole, intimidate and condition people to accept a society of total surveillance and control by your "superiors". And that's it. And it's working unfortunately, people excuse it and start to go along with it, make excuses after excuse, poof, one generation from at least semi free people to total serfdom and a fascist police state without any hassle..
There is no evidence at all that "humans in the loop" are doing anything to improve this situation, and all the real evidence points to the "humans in the loop" being good tame little heel clickers and sieg heilers.
Speaking of which, anybody know what patent numbers were issued? I can't find them in the USPTO database.
That does not compute.
You're under arrest!
Ray Davis is the founder and visionary behind the BRS technology and serves as Chairman and CEO. In 2000 Mr. Davis founded SimDesk Technologies and served as Vice-Chairman when he left in 2006 to start BRS. SimDesk created revolutionary technology that changed the very business model of the computer industry. By creating a server that can out perform any other system by a thousand times, SimDesk has been able to sell its products to federal, state, and local governments which in turn provide the software to millions of users in the U.S. and overseas. In 1995 Mr. Davis founded CyNet, Inc., an Internet related company that ranked within the top ten fastest growing companies in Houston in 1996 and 1997. Mr. Davis sold the company in 1998 and the company went public in 1999.
For the uninitiated about Simdesk and Virtudyne:
http://thedailywtf.com/Articles/Virtudyne_0x3a__The_Founding.aspx
http://www.bloghouston.net/item/6996
I've installed these systems, and they're not ready for prime time. The last customer turned off a $50,000 investment because of a 2:1 ratio of false positives annoying the operators. This bozo thinks a 3:1 ratio is going to be acceptable? Not in the real world.
"Think about how stupid the average person is. Now, realise that half of them are dumber than that." - George Carlin
People with physical and mental disabilities will be unjustly harassed by enforcers who use systems like this.
It's only a matter of time before a lawyer gets a call from someone with a limp, a tremor, a facial tic, or an itch in an unusual place and put the kibosh on the whole system. Hope they have a good legal team.
Never pulled guard duty, did you?
Actually, yes I have. You'd be surprised how "abnormal behavior" fails to appear when a guard, even unarmed, walks by every few hours.
Pavlov wouldn't be so famous if he'd used a can opener instead of a bell.
Mea culpa. I booted the description by leaving a word out. It should have been "... rapidly become the best friend of the fellow who now has to constantly scan 100 surveillance screens for unusual activity."
No mod points, no meta-moderating/Firehose/all the other free work Slashdot wants me to do.
As someone who makes a rule of behaving abnormally, I wonder how abnormal it will find having the lenses shattered at a distance greater than it's set up to see...
"Probable cause" for searches doesn't mean "my flaky security system thought you were acting suspiciously" and I hope someone immolates these guys in court the first time that's suggested.
Thankfully this kind of thing is a little way off where I live - the nearest well known "1984" surveillance state is a bit further west in Kelowna, BC. I still plan to fight this stuff tooth and nail where I can though, as so far it's only been abused and of questionable value in deterring crime, but it's definitely encroaching on remaining free cities.
I work with machine vision and cameras like these, and I have a feeling that by "objects" they mean that they've taken an image of the background and "diff"d it from the current image, then they can analyze the various objects that remain (are they connected? What color are they? are they in the same 3-d plane? Do they have the same recurring textures? etc.)
All these things are easy to do, and in fact, I have software on my computer right now (in the form of HALCON dev studio examples), that tracks cars coming and going from an intersection, people walking along the sidewalk and through the woods, and various other moving objects. I count 39 moving objects (including a tree that is swaying).
What this really is, is another frivolous patent, because I bet these examples are now infringing that patent. I guarantee that nothing they're doing is new.
There's already better, faster systems out there. I work for a remote monitoring station that monitors hundreds of -sites- remotely. We use a combination of software/hardware, but the best system we have going right now uses this http://www.aimetis.com/symphony/default.aspx It does everything the article claims the new stuff does, without the 'learning' aspect, which is essentially useless anyways, as there is no 100% predictability of 'normal' activity.
Don't have to go in fer a squirrel suit. I had a camera in my basement that was left on at night simply because I was too f#@king lazy to turn it off. It usually would wind up with several recordings of disturbances in the air that looked for all the world like GHOSTS! Zounds! Now is similar 'spirits' got into their 'digital systems', maybe 'Tron' will live again!