Slashdot Mirror
Video Surveillance System That Reasons Like a Human
An anonymous reader writes "BRS Labs has created a technology it calls Behavioral Analytics which uses cognitive reasoning, much like the human brain, to process visual data and to identify criminal and terroristic activities. Built on a framework of cognitive learning engines and computer vision, AISight, provides an automated and scalable surveillance solution that analyzes behavioral patterns, activities and scene content without the need for human training, setup, or programming."
Nothing can go wrong!
Source or it doesn't work.
A little more info from the BRS Labs website:
...
but still send an alarm when a
burgler steals your stuff!
"The system takes the input from existing video security cameras (no need to change equipment); recognizes and identifies the objects in each frame and passes that data to its Machine Learning Engine. There, the system 'learns' what activity is normal for each unique area viewed by each camera. It then stores these LEARNED memories, much the same way the human brain does, and refers back to them with any and all future activities observed by the camera. If any behavior falls outside of the norm, alerts are generated."
Sounds impressive, but will the algorithms be sophisticated enough to watch grass grow and realize that it's normal behavior for the garbage truck to come by weekly
Hulk SMASH Celiac Disease
that was a press release for the company's product. It has no reliable or interesting information whatsoever.
So I guess this means that the camera is going to Harass people taking Photos now?
It's a press release pretending to be journalism.
If it doesn't need training, how does it define "terroristic activity"? Is it the "I'll know it when I see it" definition?
The article seems to indicate it works like a Bayesian filter on the video - pointing out things that aren't typical for the camera.
Much like any automated system that is supposed to filter out false positives, it is probably pretty easy to train either the operators or the system itself to throttle back the sensitivity to a point where it ignores everything.
...that somewhere else in the world, there is a young, badass mother fighting off robots from the future that were designed to look like my Governor in a heroic attempt to destroy this new technology along with her scrappy, but as-of-yet slightly immature son....
At least, I think that's where we are in the time-line right?
Motorcycles, Robots, Space Gossip and More!
The "machine learning engine" is a "datacenter" (warehouse) full of cheap African laborers who are all watching the cameras.
(this is a joke, it just isn't funny, and it is meant to illustrate a point. See the next line):
God/nature/FSM/evolution/al gore/$deity has done a pretty damn good job at building our brains, why are we trying to reinvent that wheel in a computer?
NewslilySocial News. No lolcats allowed.
Eagle Eye is not a blueprint for your surviellence computers. Thanks.
After logging in slashdot still does not take you back to the page you were on. It's been that way for 20 years.
...does it run racial profiling?
THL phish sticks
So if it watches Gary Indiana, murder and mayhem will be programmed in as normal?
---- Booth was a patriot ----
Ocean's Thirteen that a system like that.
What a great way to absolve any personal responsibility. Detained wrongfully? Not our fault, the machine said you were moving like a terrorist.
How is this fair to those who have a complex to appear to act criminally, though they are trying their hardest to act normal and not commit crime?
Human judgment isn't accurate enough to distinguish between an actual terrorist and someone who may look like one. Why is there anyone expecting good results from a machine emulating this judgment that isn't reliable in the first place?
Much like detecting terrorists by facial recognition, this is vaporware until they publish some numbers.
I once had someone misplace a sales call to me, being proud his facial recognition system was 70% accurate. He had no idea how much his system is a pain in the ass when its wrong, and for the airport security business he was trying to get, 90% accuracy is considered terrible.
Thank God it reasons like a human does. As we know, humans have a 0% false positive rate at identifying potential terrorists. Now, not only can we identify old ladies in wheelchairs with oxygen tanks as the terrorists they are in a completely automated fashion, but we can do it at speeds previously only dreamt of!
Maybe they shouldn't have used human cops as their behavior model after all.
Surveillance System becomes self-aware, becomes as "efficient" as any other TSA at spotting terrorists, realizes the futility of security theater, quits in disgust, writes a best-selling autobiography.
Really I am sick and tired of the surveillance realm. If anybody really wants to do something nefarious they will make sure the cameras don't work. Simply pull them down, spray them with paint or whatever. The authorities will not come running. After the fact usage is good, but really it doesn't stop any crime, even the random ones.We are the ones funding this and do not even have a say in it.
From the description and providing it is smart enough to identify the race of an individual, it will sound an alert if a white person enters a monitored area normally only inhabited by non-whites and vice-versa. Gee now we're advanced enough to build our racism into our computers. Wonderful
This is just one more step along the path that I and many others have predicted. Please reference the thread on "crowdsourcing" video analysis: http://slashdot.org/comments.pl?sid=1277749&cid=28429931
The whole point of this is:
1. Sell to customers who blindly trust in it.
2. Fail to detect anything on many an occasion because it most likely isn't perfect.
3. ???
4. PROFIT!!!
"Our country is not nearly so overrun with the bigoted as it is overrun with the broadminded." -Archbishop Fulton Sheen
The central processor in this thing isn't a computer at all... it's a dead salmon!
I wonder if it's based on Numenta's Bayesian HTM (hierarchical temporal memory). My understanding of neuro-like learning system is that, unless its knowledge base is organized hierachically like a tree, it could not possibly do the things its promoters are claiming for it.
xkcd, source, video, pictures, audio, podcast or it didn't happen.
So, it instinctively directs the cameras towards the hot women all the time, distracted from important things it should look at?
Interestingly in Europe after a series of dreadful incidents on live video, this is finally being debated on the eve of general elections: http://www.piratenpartei.de/node/920/29268#comment-29268 - as at the other end of the line, in a situation room (that may be on the next floor or station, and yet too) far away, officers will have to watch events unfold and wish in vain to finally be out there with a gun again (or have sufficient forces to dispatch), e.g. to stop that attacker they can only videotape and helplessly watch wreak havoc on screen.
I can just see the kids in the UK figuring out what kind of innocent activity triggers police reactions. When the flood of false-positives starts, the cameras will be back to being as use[ful|less] as they are today.
ON DELETE CASCADE
What are the odds these cameras won't be able to distinguish between people fighting and people shagging?
Slashdot: news for Apple. Stuff that Apple.
"That Reasons Like a Human"
Spends most of it's time ogling - wait there's a honky in this black neighborhood, must be up to no good.
Nate
It will lie.
For justice, we must go to Don Corleone
to comply!
No one talkd about ED-209 yet?
What time is it/will be over there? Check with my iPhone app!
Seriously, I would love to see some of this surveillance tech turned back on the government. Install this thing in congress and train it to watch for corruption. It would probably fill up a massive disk array in a couple hours with positive hits.
you give us another billion dollars to finish it.
Yeah, right.
A 1% error rate will produce a hundred times as many false positives - all innocent people accused of a crime - as real positives. And a 20% error rate is far, far more likely.
Scams like this are the reason why you have to show up at airports three hours early now.
Is it smart enough to knwo that "terroristic" isn't a real word, at least?
hmm.. actually, this idea is great! except for one bit of a problem.. this will have a profound effect in future human behavior. since criminal acts/behavior would have to change incredibly fast to keep beating this 'sauron'-bot, every 'illegal behavior' might become extinct.. is that really a good thing given the laws we have today?
This system does not operate like a human at all. A human operator does not look for signs of terrorist activities. A human operator looks at boobs.
Equine Mammals Are Considerably Smaller
So, it makes wild guesses, allows others to tell it how it should be thinking and/or bases vital decisions on obviously false beliefs?
I didn't RTFA or This one, but it looks similar.
It's just another bloated pentagon pork project of no real value or merit. We see this all the time. It reminds me of avant garde art, only 10,000x more expensive and twice as pointless. But only half as ugly.
RS
Shoes for Industry. Shoes for the Dead.
a) Terroristic? Not just terrorist activities? /is/ usually a crime, nothing special.
b) Terrorism
Colossus: The Forbin Project
So how long before this thing figures out how to pork a co-worker on lunch break, record the act on one of the cameras it's supposed to be monitoring, and piss in the boss's coffee?
I'm betting about three weeks.
I've calculated my velocity with such exquisite precision that I have no idea where I am.
it's great that it will alert you to possible terrorist activity, but will it also alert you to female nudity?
Fire all the cops and judges, convert them all to prison guards, and we'll make the city a jail.
So basically a "red light camera" for people.
And like the red light cameras, there's no way to appeal to human judgement if the camera says you're guilty, you must be guilty unless you can prove you are innocent (for red light cameras, at least in California, that means proving the amber light lasted less than 4.8 seconds).
I love the presumption of guilt they're slowly building into the system in the name of revenue generation. "The war on terror" has been going on for 8 years now, and they finally arrested _a_ suspect with no concrete plans of timing, location, or a target. I will bet that this system will ultimately be used to automatically issue jay-walking tickets.
-- Terry
...what sort of fun we could have if this system was used as the eyes and brains of an armed security robot?
It will get bored and fall asleep. I can't wait to see the R&D explain this one: "Well, at 9h35mn PM, the image became out of focus and aimed at the floor under the camera..."
It will look around for interesting "sights". "Yes sir, the system was fixing on this lady, we deduced that this new camera was bogus and probably a terrorist weapon of some sort..."
There is one rule to remember here: Humans are best at behaving like humans.
"God/nature/FSM/evolution/al gore/$deity has done a pretty damn good job at building our brains, why are we trying to reinvent that wheel in a computer?"
Because nobody like shitty, repetitive, unfulfilled, degrading work. Now I recommend you look up definition of "Weak AI". You may be able to find it in encyclopedia in a library around you or may be in your house if for some reason you disapprove of not evil corporation that shall remain nameless forcing hundreds of teraflops worth of enslaved magical pixies (that is a lot of pixies, oh my) locked in "servers" (steel boxes) to do the thinking for you.
To do what AISight does one needs:
Moving objects must be dynamically tracked and their behavior somehow segmented into steps. This is arbitrary in that there can be a theoretically infinite variety of such segments for any given macro behavior. E.g., I can say "Mary handed the book to Tom." or I can say "Mary grasped the book, Mary extended her arm in the direction of Tom, Tom extended his hand in the direction of Mary and grasped the book, Mary un-grasped the book, Tom retracted his arm (with the book)." Both describe the same action, but the second has a finer segmentation of behavior.
In summary I don't see that there is much new here: certainly all the pieces are available. It's essentially the classic AI expert system. Since they're getting millions for known technology (and software), I expect to see a freeware version of this available within the next 6 months!8-))
To do what AISight does one needs:
Moving objects must be dynamically tracked and their behavior somehow segmented into steps. This is arbitrary in that there can be a theoretically infinite variety of such segments for any given macro behavior. E.g., I can say "Mary handed the book to Tom." or I can say "Mary grasped the book, Mary extended her arm in the direction of Tom, Tom extended his hand in the direction of Mary and grasped the book, Mary un-grasped the book, Tom retracted his arm (with the book)." Both describe the same action, but the second has a finer segmentation of behavior.
All the pieces are available in freeware. It's essentially a classic AI expert system. Since they're getting millions for known technology (and software), I expect to see a freeware version of this available soon!8-))
... because even surveillance is not purely objective. Think about selling it to Islamic countries, for example. Suddenly drinking alcoholic beverages should raise an alert, marrying a second, third or fourth wife must not. There is a whole new market for Hijab, the head-scarves, scantily clad woman. But even in the Western society, a lot needs to be learned. Finns carry their wifes one day in a year, while a woman carried around by a man is 'offensive' in any other society.
Yes, man, there are absolutely zero objective standards for what a surveillance system should and should not detect. May I myself climb my fence when I lost the gate key? How does the system know? Or does it shoot me on preemptive reasoning?
BRS Labs was founded by Ray Davis (http://www.brslabs.com/company_overview.php) who formerly founded SimDesk. SimDesk was the MS Office knockoff that plowed through hundreds of millions before shutting down operations in 2008. SimDesk also made outlandish claims about its software, such as claiming that it could handle "millions of users on a single server" (http://www.usatoday.com/money/industries/technology/2003-01-21-cover-side_x.htm).
Install this thing in congress and train it to watch for corruption. It would probably fill up a massive disk array in a couple hours with positive hits.
I understand it can only detect abnormal events. Now in many parliaments wouldn't that mean: alert to exceptionally rare cases of non-corruption? You know Professor Lessig changed his focus of research for a reason.
Whatever the scene may look like to an "intelligent" camera, "Lobbyist walks into lawmakers' offices and leaves without the two black suitcases s/he brought" is probably not an instance of someone planting bombs (at least unless the pictures make it to the pages of the Post).
To quote Ambrose Bierce:
politics, n.: The conduct of public affairs for private advantage.
(Others have explained the word as a composite of poly and ticks...)
Or as Bismarck once said:
The people sleep better at night if they do not know how laws and sausages are made.
Now they just need to merge this with a federally required RFID marker to identify you and the system can simply flag you as a threat automatically whenever you act outside of set threshold determined by a collectively gathered log of your previous day to day activities.
8==8 Bones 8==8
I read the article. I've been the sad sap (with three to four others per shift) in the alarm center touring 80 on-site and 600 remote cams with associated door/gate/motion/fire sensors.
I've also been the sad sap moving mice from dirty cages to clean cages. No one has been able to automate that. Wish I could, because that patent would be worth hundreds of millions, easily.
The notion that you're going to replace a vigilant PTZ operator who knows the site and the flow with a program is ludicrous. A program will never be a brain.
Granted, some programs work better than some brains, but that's a different topic...
This is much the same concept as recently presented at ICDSC-2009 in the paper "Abnormal Motion Detection in a Real-time Smart Camera System", based on real-time video analytics and artificial neural networks to autonomously build a model of what is considered normal behavior, subsequently flagging outliers as possible events that require further (human) scrutiny. Thus it acts as an intelligent self-learning classifier/filter that greatly reduces the information stream per camera for human operators.
This would probably trigger on me, because I have a habbit of always acting strange in public.
When I am in a large crowd and wants to move quickly, walking fast is too slow, but running is difficult due to quick changes in momentum and constant need for acceleration. I have a special technique where I am constantly jumping and alternating between two steps on the left and two steps on the right leg. This allows me to move quickly and maintain speed and quickly pick up acceleration when momentum changes, and allows me to quickly stop and prevent collision. It is a very effective way of moving, although it looks kinda strange.
While waiting for the train to arrive, or while I am on the train, I also tend to dance. To pass time, have fun and to improve my dancing skills.
On the train, there is a bar hanging from the roof, which people can hold on to. I use this bar to do pull-up/chin-up exercises. It gives me something to do, makes use of my time so I don't waste it, and gives me healthy training and exercise.
Sounds like the touts in Saigon, trying to sell us their package tours to the very touristic beaches. (No offence meant to the Vietnamese).
Seriously, ever heard of a dictionary?
(the captcha: seconds: only too appropriate).
that's funny, because even young human beings need to learn "good" from "bad", under adult supervision, it's not an automatic process.
It can figure out that Little Tiffany is dangerous and deserved to die?
Many theories, but few empirical facts.
Oh yay, I wonder that would react to my new hobby... or to me in general =/