US Spends $11M To Kick-Start Video Search

coondoggie writes "The US military is inundated with video from airborne unmanned aircraft, remote monitoring systems and security outposts. In an effort to speed up the processing and analyzing of all this video, researchers at Defense Advanced Research Projects Agency (DARPA) this week awarded an almost $11 million contract to open source software vendor Kitware to help develop what DARPA calls its Video and Image Retrieval and Analysis Tool (VIRAT) program."

Comments

Are they really that Open?

[stanlyb]

Kitware, beware, there is a saying, do not accept even the gifts from Danains.......

Re:Are they really that Open?

[oiron]

In case you didn't know, kitware's clients already include Sandia labs, and various other agencies.

They still keep the core products (VTK, ITK, CMake, etc) open, but will build custom solutions on top of them for clients.

The gift in question isn't really that much different from their usual work...

lulz, acronyms

[Pojut]

ATGTGAATTPRTSOIC*

*Are they going to give an acronym to the people running the software once it's complete?

Re:lulz, acronyms

[Anonymous Coward]

That's an initialism, not an acronym. An acronym is something you can pronounce.

Re:lulz, acronyms

[hedwards]

Smart ass, it happens to be pronounced "Fred."

Re:lulz, acronyms

[Anonymous Coward]

So, Artgotgaatotpertsoonic is better?

Re:lulz, acronyms

[nomorecwrd]

Much, much better.... specially now that reads like a common, everyday, word.

Re:lulz, acronyms

[Anonymous Coward]

An initialism is a kind of acronym.

Initialism is a word used exclusively by linguistics scholars and aspies^H^H^H^H^H^Hwikipedians.

Just as it's common to refer to NaCl as 'salt' when in fact it is only one of many different salts, it's common to refer to initialisms as acronyms.

Don't be a douche.

Re:lulz, acronyms

[Anonymous Coward]

actually, i think VIRAT is a pretty good acronym. it could also stand for Video Intelligence RAT system. i'm sure it will increase their ability to rat on people.

ha! and my captcha phrase for this post was 'dictator' ha!

not enough money

[M. Kristopeit]

for the level of processing required, $11M will barely cover the hardware infrastructure

Re:not enough money

[MozeeToby]

It's a DARPA project, they are almost by definition proof of concept, ten years out, 'most likely won't work this time around (but hey, wouldn't it be cool if it did?)' type projects that are designed to get the ball rolling on technology that might be possible to implement today. No one in charge of this project is expecting to roll it out into combat situations next year, they just want to see what a bit of money thrown at the problem comes up with; they literally don't even care if it's successful or not.

Re:not enough money

[Anonymous Coward]

It's an insult for the military to be using tax dollars to work on developing Computer Vision technology 20 years ahead of it's time when we're up against record unemployment.

Burn them to fucking DVDs and hire analysts to work from home dipshit. The results you get from a McDonalds employee are going to be 1000 times better than any computer in the next 100 years.

If they're feeling particularly ambitious they can profit share/crowd source. Then you'll give a bunch of retired military grunts a hobby where they can contribute to the fight and help pay for their retirement.

If they just want a "YouTube" for classified video: why don't they just hire google to build them a "Youtube" for classified video?

Even better: hire more Drone pilots and give them time to properly document the intelligence they gather. That way: it's tagged at the source instead of crowd source guess work.

Fuck the military for being so naive that they think this is a problem they can throw computers at.

This is coming from a robotics nerd btw.

Re:not enough money

[Anonymous Coward]

Ha! I get the troll! Very funny -- hire McDonald's employees to analyze military secrets. Hoo hoo, obvious jokes are the funny.

Re:not enough money

[blackraven14250]

It's an insult for the military to be using tax dollars to work on developing Computer Vision technology 20 years ahead of it's time when we're up against record unemployment.

"It's an insult for DARPA to create jobs at Kitware during 'record unemployment' that isn't actually record unemployment, just high unemployment."

-AC

Re:not enough money

[M. Kristopeit]

since when does youtube allow you to search for objects in videos? youtube search works off keywords and the title and description of the video... it's pretty obvious you don't understand what this new project is intending to do.

Re:not enough money

[stewbacca]

I actually laughed out loud when I read $11m. All I could picture was Dr. Evil...."11 MIIIIILION Dollars"...then a bunch of people laughing.

Seriously, my crew of 4 costs the government 10 times that amount for glorified PowerPoint training slides.

Re:not enough money

[DeadboltX]

How do I get my boss to adapt this business strategy?

Re:not enough money

[hedwards]

Depends how you measure it. There's probably more people out of work now than there were during the great depression, even though the percentage is lower. The last completed census put the US population at ~281m, whereas the 1930 census put the US population at ~123m, so while it's hard to say with any certainty, it's definitely possible that there are more people out of work now than there were then. Plus we haven't got any WPA style programs going to help those out of work earn their keep even as we do little to fix the problems.

Re:not enough money

[Anonymous Coward]

Brilliant insight asshole. Lets hire 1000 skilled laborers to spend 10 years on speculative R&D which may or may not bear fruit.

Where will we get the money to do this?
From taxes of course! Taxes paid for in part by the 10 percent of the population who is unemployed right now: unskilled laborers.

What valuable work will those 1000 lucky individuals be applying themselves to you ask?
Making obsolete one of the few jobs left that unskilled humans are better at than machines!

Why is this such a dumb fucking idea?
A. Because the private sector is already working on it at a pace appropriate to the cost of processing power.
B. Because humans are so much cheaper than super computers.
C. Because while one hand of the government:
-taxes jobs out of existence to finance ridiculous computer vision toys for the military which are better filled with humans at a time when there is a literal sea of unskilled labor
the other hand:
-"invests" money in "jobs programs" like repaving every fucking inch of road in the country, while building maintenance and bridge repair continue to be neglected.
D. All of the above.

Our entire fucking economy is just one giant blender designed for the sole purpose of making tribute to the "broken window fallacy."

The only fucking time the government get's it right and fights back against feather bedding and make work: it's to burn down the single most obvious opportunity to take some of those greenhouse employees displaced in the rush to destroy window panes, and put them back to work doing something productive.

You cannot make this shit up.

Re:not enough money

[MozeeToby]

Just ask your boss if you can have 0.48% of the company's budget for high risk, high payoff projects (DARPA gets $3.2 billion of the military's $660 billion budget) . I'd be willing to bet that most engineering places put more than that into advanced prototyping and proof of concept designs as it is. Doesn't seem so bad when you put it that way does it?

Re:not enough money

[blackraven14250]

Brilliant insight asshole. Lets hire 1000 skilled laborers to spend 10 years on speculative R&D which may or may not bear fruit.

Quite frankly, who gives a shit if it bears fruit? Every bit of R&D ever done has been done under the pretense that there's the possibility of the project failing, whether it's a very likely chance during research, or a moderately to slightly likely chance during development.

1000 people definitively employed for 10 years, or 1000 people who may or may not get a job in the next 10 years. Your pick.

Re:not enough money

[Anonymous Coward]

I'll take 50,000 doing this job better than a computer for the next 10 years for the same price tag.

I forgot the philosophy of government:"Why buy 1 when you can buy 2 for twice the price!"

I'm sure there's a literal ton of Chair Force analysts getting paid 100X more than this job is worth because they passed the super challenging process of a lie detector test & their neighbors not having anything nasty to say about them.

This shit should be crowd sourced. Fuck government secrecy.

Re:not enough money

[blackraven14250]

Unemployment is typically measured as a percentage, not as a raw number, since the relative scale of unemployment is of real importance, not just how many people. 500k people unemployed in the US is amazingly low, but that number would be crippling to Bulgaria.

But, assuming we're using the raw, non-standard measure, unemployment peaked at around 22.5% during the depression, a lowish estimate. 123m * 21.5% = 27.67m

We currently have 9.5% unemployment (July 2010). 281m * 9.5% = 26.695m

Considering official depression-era unemployment statistics don't exist, that's an estimate up there, so the number could be decently different. I also didn't do any research into census numbers; I'm assuming yours are accurate.

Re:not enough money

[Anonymous Coward]

There's also the demographic differences. In the Depression Era, mainly men worked. So, out of a population of 123m, you have a workforce of mostly men, which would have roughly been 25% of the population. In today's economy, both men and women work so the percentage of the population that's actually considered part of the work force is much larger. However, it is not so much of an option that both men and women work as it is more of a necessity as steady inflation has pretty much required the presence of two-income households. The effects of losing just one job today can magnified and also buffered by the two-income household requirement; depends upon each particular case. We may be experiencing less unemployment, but it may feel like the same unemployment to some. And the U-6 rate is near 20% (which is a reflection of how it 'feels'), so don't go solely by the 9% published U-3 rate.

Re:not enough money

[drerwk]

What do you do that costs $25M for an FTE?

Re:not enough money

[jpapon]

You won't need supercomputers to do this, it will likely be done on massively parallel processors, probably using CUDA. At least that's how I would do it.

Are you against all progress, or just in this particular instance? Because you sound like you'd be arguing against developing tractors because they'd take away jobs from farmers.

Or developing steam ships because you wouldn't need as many sailors.

Or wheels because they'd take away the jobs of litter-bearers.

Or fire because you wouldn't need lots of people to snuggle with at night to keep you warm.

If people like you got their way, we'd still be hunter-gatherers on the plains of Africa. But guess what, even then, there'd still be people like you bitching about the people trying to do something useful, like make a spear to hunt with.

Re:not enough money

[Nikker]

These people aren't stupid in the slightest sense of the word. The offer one prize of 11M which if there are quality people with quality hardware isn't really that much but the catch is I have independent groups of people all doing research and finding many different and unique ways of doing it. The best part is each group has to show me what they have done and I only have to pay once for everyones work! So instead of hiring NASA to come up with the solution for 100M+, I can have 10 or 20 groups doing R&D for price of 1. Sounds like Profit!!!

Re:not enough money

[totally bogus dude]

Problem 1: security clearances. Much of this information is classified, and some of it deservedly so. The more low-skilled people you have looking at it, the more likely you'll have leaks and other problems.

Problem 2: the job is probably harder than you think. Many people will give suboptimal results due to laziness or ineptitude. You have to train people, somehow test they're still performing well, give incentives, weed out the bad ones, and all the HR type stuff as well.

Problem 3: it's a tedious and boring job. People tend to get blase when doing tedious boring jobs. Consider the cliche of the sleeping security guard, or the donut-eating police officer. Most of the time, these videos will be b.o.r.i.n.g. The low-skilled analysts will get sloppy.

Problem 4: what's deemed "interesting" may change. An analyst today might be looking for particular types of sedans, since it's currently believed the people we're looking for use that kind of vehicles. In three months time, intelligence might emerge that they actually drive vans. Now you have to have your analysts go back and re-analyze all the videos from that period looking for vans instead. Either that, or you make them tag every single thing in every single video, just in case. Massive tedium -> massive amount of errors.

Computers on the other hand, are very good at doing mind-numbing tasks hour after hour with absolutely unwavering attention to detail and consistent quality of results. A relatively small number of human analysts can (and I'm sure, will) be employed to allow the quality of the computer's output to be gauged. As improvements in the computer's algorithms are made, previous videos can be re-processed to take advantage of that.

Also, I don't think the costs make much sense either. A few hundred thousand dollars can buy a massively powerful collection of commodity computers, and this is a task which inherently lends itself to parallel processing. A few hundred thousand dollars would pay the salaries of the army of low-wage, disinterested "analysts" for a year at best.

Finally, none of this rules out using humans to also analyze videos, particularly the subset of videos that are likely to contain useful intel.

I have seen those videos...

[Anonymous Coward]

DarpaVid Search: "Brown People" (showing 10 out of approx. 1,342,400,000 video results)

DarpaVid Search: "Brown People +terrorist" (showing 10 out of approx. 2,670,000 video results)

DarpaVid Search: "Brown People +terrorist +diabetic +tall" (showing 10 out of approx. 4,000 video results)

DarpaVid Search: "Brown People +terrorist +diabetic +tall +scratchybeard +inexplicablycleanclothes" (showing 1 out of 1 video result)

AHA! WE GOT HIM!

Great, another privacy kill

[gurps_npc]

The main remaining limit that prevents people from making real use of the enormous number of vid cameras in the world to track and invade our privacy is the fact that we have an enormous number of vid cameras. You basically need a person for every 10 cameras or so and watching it is boring.

Once someone comes up witha way to automate the scanning of all that video, then our privacy will always be gone, as opposed to simply invaded after a crime is committed.

Re:Great, another privacy kill

[tehtest]

If you are in a public space, by definition, you have no privacy. I'm so fucking tired of people crying "What about my Privacy!". If you would like to maintain your privacy, stay in a physical location which is private. How the crap can you expect to maintain privacy in PUBLIC?????????

Re:Great, another privacy kill

[hairyfeet]

Think you are missing the point dude, the point is power. look up something like COINTELPRO and see how they were able to pretty much derail many groups just by keeping tabs on the right people. But all that takes manpower, now imagine: what if it didn't? What if you could dig up dirt on any politician, police officer, anybody you wanted, simply by "flagging" his likeness and having the cameras spread out all over their route do the work for you? Or have it go through all the archives and auto flag anything they've ever done? Want that new money sink bomber that senator asshole won't vote for? Hey senator, here is video of you going into a hotel room with your page! Here is you staggering drunk 3 years ago!

The danger isn't about privacy, it is about power. Have you EVER done anything in public that could put you in a bad light, or even be misconstrued to make you look bad? Well then you better play ball sonny, or it will be on the latest tabloid by tomorrow. Now THAT is the danger we are looking at with this kind of crap.

Re:Great, another privacy kill

[tehcyder]

Once someone comes up witha way to automate the scanning of all that video, then our privacy will always be gone, as opposed to simply invaded after a crime is committed.

If the police weren't allowed some measure of invasion of privacy, no crime would ever be solved, as even a full confession could be judged to be self-invasion.

I suspect google is already doing a lot of this

[peter303]

With their multi-billion investment in youtube. They need to find the good stuff in there. All five seconds of it! :-)

Security camera firms have similar issue. My grocery store has over 50 cameras because they are so cheap. But I doubt they have the eyeballs to view a small fraction of it.

Re:I suspect google is already doing a lot of this

[MozeeToby]

I'm just guessing, but I have worked a little bit with security feed monitoring software before. Most likely they have 4-8 key cameras (on the registers and liquor department) which are shown 2-4 at a time and rotated through to watch for actual theft or violent behavior. Around these, they will have a number (10+) of lower priority feeds being displayed scaled down and rotated through more slowly which basically only serve the purpose of watching for weird, suspicious behavior.

The rest of the feeds probably aren't even watched but digital storage is cheap these days, it's pretty trivial to keep a decent quality recording even of 50 feeds for the past day or so, with a simple panic button to prevent deletion if something happens. Depending on how elaborate the system is, there might also be a way to flag feeds that meet certain criteria and display them in the main displays with an alert. Things like motion in what should be an empty stockroom, people moving backwards through the registers, fire doors opening, etc.

Of course, the even more likely answer to the issue of having 50 'cameras' and no one to watch them is that there are really only 5 cameras and 45 opaque plastic domes that look like cameras. That is the solution that the vast majority of stores choose to go with.

Re:I suspect google is already doing a lot of this

[Stephenmg]

Most places don't install cameras to prevent bad behavior or even to catch it in the act. The real point of most video surveillance is to tell that bimbo that said he/she fell in isle 3 he/she is full of it. I work at a place with about 200 cameras in about 15 buildings. No one actively watches them, we just get request for video of a location at certain time and burn it to a CD.

Re:I suspect google is already doing a lot of this

[Anonymous Coward]

Actually most cameras are used to catch and document employee left. The loss from actual shoplifting and slip-n-falls is trivial to the ammount stolen by employees. I say this as someone who does surveillance work for a living.

Anon for obvious reason.

Kitware?

[guruevi]

That company is not truly open source. The only thing they do distribute is VTK (which is only a library under a BSD license) but programs that are actually useful without being a programmer (as most researchers that use this stuff are not programmers) are under a more restrictive license. It's similar to saying Mac OS X is open source because the kernel is open or Windows is open source because BIND and the TCP stack is derived from BSD. They use the open source label to get the community to fix their problems on some of their other stuff but the non-commercial releases are fairly unstable or plain useless.

Re:Kitware?

[Anonymous Coward]

I think you forgot CMake, a pretty significant open source tool distributed by Kitware.

Re:Kitware?

[seekthirst]

Complete nonsense: ParaView, ITK, CMake, CDash, Slicer, Titan, MIDAS, vxl, IGSTK, and more: all open source, some are toolkits, some are applications. These tools are in widespread use in production environments. The company teaches an open source course at RPI, and particpates in things like OSCON, etc.

Re:Kitware?

[Paul Fernhout]

They have other things. MIDAS, for example, is really spiffy and under a BSD-ish license, and is probably part of why they got this contract:
    http://www.kitware.com/MIDAS/resources/software.html
"MIDAS integrates multimedia server technology with Kitware's open-source data analysis and visualization clients. The server follows open standards for data storage, access and harvesting. MIDAS has been optimized for storing massive collections of scientific data and related metadata and reports. MIDAS is available under a non-restrictive (BSD) open-source license."

More on the sorts of issues any FOSS-oriented progressive company may struggle with (by me):
    "Re: [Open Manufacturing] Open source manufacturing social organization"
        http://groups.google.com/group/openmanufacturing/msg/6819187b74f4b7db
        http://groups.google.com/group/openmanufacturing/msg/fa4459793c6b7ed3
    "Jobs at Materialise 3D in the Ukraine; thoughts on social change"
        http://groups.google.com/group/openmanufacturing/msg/04fbdf60ad463dbb
    "Beyond a Jobless Recovery: A heterodox perspective on 21st century economics"
        http://knol.google.com/k/paul-d-fernhout/beyond-a-jobless-recovery

But I might just be saying this because I live not too far from them and maybe I'll need a job there someday. :-) And they might have the contacts and social infrastructure to get this project better funded: :-)
    "The need for open source sensemaking tools (Score:5, Interesting)"
    http://slashdot.org/comments.pl?sid=1746980&cid=33177866

Re:Kitware?

[Paul Fernhout]

Oops, the first and third Google groups links are swapped. I didn't check those URLs carefully enough before posting. Dang, there goes my chance to impress them as a careful programmer. :-)

Re:Kitware?

[Paul Fernhout]

And of course, as I posted elsewhere in this discussion, the whole project itself is still pretty ironic:
    http://www.pdfernhout.net/recognizing-irony-is-a-key-to-transcending-militarism.html
"Military robots like drones are ironic because they are created essentially to force humans to work like robots in an industrialized social order. Why not just create industrial robots to do the work instead? ... Still, we must accept that there is nothing wrong with wanting some security. The issue is how we go about it in a non-ironic way that works for everyone. ..."

Re:Kitware?

[Anonymous Coward]

Must be nice only

Re:Kitware?

[oiron]

Open source != "I don't like it".

been here before

[recharged95]

Remember when we tried searching video aka Oracle VIR? I sure do (part of a firm that used VIR as a core for a NL video search engine).
Yes, 11mil is not gonna cut it.

On the irony of military robotics...

[Paul Fernhout]

My comments: http://www.pdfernhout.net/recognizing-irony-is-a-key-to-transcending-militarism.html
"... Likewise, even United States three-letter agencies like the NSA and the CIA, as well as their foreign counterparts, are becoming ironic institutions in many ways. Despite probably having more computing power per square foot than any other place in the world, they seem not to have thought much about the implications of all that computer power and organized information to transform the world into a place of abundance for all. Cheap computing makes possible just about cheap everything else, as does the ability to make better designs through shared computing. ... There is a fundamental mismatch between 21st century reality and 20th century security thinking. Those "security" agencies are using those tools of abundance, cooperation, and sharing mainly from a mindset of scarcity, competition, and secrecy. Given the power of 21st century technology as an amplifier (including as weapons of mass destruction), a scarcity-based approach to using such technology ultimately is just making us all insecure. Such powerful technologies of abundance, designed, organized, and used from a mindset of scarcity could well ironically doom us all whether through military robots, nukes, plagues, propaganda, or whatever else... Or alternatively, as Bucky Fuller and others have suggested, we could use such technologies to build a world that is abundant and secure for all. ..."

Re:On the irony of military robotics...

[Simulant]

At this point I'm pretty sure that one day geneticists will discover that there is a gene for irony which at least half of our population lacks. :(

Slashdot reaches new degree of separation from new

[Myopic]

We are sitting here commenting on a Slashdot blog post, which links to a Techdirt blog post, which links to a blogs.journalism.co.uk blog post, which links to this news article.

I skipped the blogs and read the article.

Ever notice

[SnarfQuest]

Ever notice that all the major inovations on the internet are porn related?

Why not outsource that job?

[Arancaytar]

I'm betting Wikileaks would do it for free!

Re:Slashdot reaches new degree of separation from

[MagicM]

You should have read the Slashdot post too, since your news article refers to this post instead of the current one.

Well ...

[Anonymous Coward]

Didn't the Chatroulette people already lay the groundwork with their "penis recognition filter" project?

Here at DARPA....

[ksandom]

We like to do things organically. Now we want to search video. It's inspired by a rat, and a virus. We call it.... the VIRAT!

Might have another intended use

[Burz]

...like flagging videos of egregious and potentially embarrassing acts withing military archives...

It turns out manual labor is not a growth industry

[tlambert]

Plus we haven't got any WPA style programs going to help those out of work earn their keep

It turns out manual labor is not a growth industry.

Most of the WPA and similar projects built too well, and they are still around and in active use. FDR's "alphabet soup" of the depression era "New Deal" was predominantly manual labor, and did things (the CCC) like turning 84,400,000 of wetlands into farmland . Other major agencies were the PWA, CWA, and TVA. We're currently tearing down a number of PWA dams because of their threat to salmon breeding. The CWA cost $1B over the 5 months it existed (that's $1B 1933 dollars). The TVA is still around, as the largest public power company in the U.S, and are the single largest operator of dirty coal-fired power plants outside of China.

-- Terry

Telescreens are coming to your city

[Pomslo]

At least, in the Orwellian regime, you could be sure they werent watching *every* telescreen.It was more like not wanting to take the chance, they were looking at you. I predict, that by means of a few years we will be thinking twice about or facial and general bodily explression, fearful that someone somwhere might be laying judgement upon us. On the bright side, we wont really need an all knowing godly watcher to instill fear upon the masses, bye religion!, welcome science!

Will it involve multitouch screens?

[Psaakyrn]

(like Minority Report?)

Lost opportunity...

[msauve]

"Video and Image Retrieval and Analysis Tool (VIRAT) program"

They _could_ have called it the Bitchin' Optical Retrieval and Analysis Tool (BORAT).

Re:Lost opportunity...

[tehcyder]

Or the "Coded United Nations Terrorist Retrieval Analysis Tool"

Re:Slashdot reaches new degree of separation from

[WML MUNSON]

We are sitting here commenting on a Slashdot blog post, which links to a Techdirt blog post, which links to a blogs.journalism.co.uk blog post, which links to this news article.

I skipped the blogs and read the article.

Whoops! :/

That link is related to the recent Wikileaks/Shield Law post. Try this one instead: http://www.theregister.co.uk/2010/09/01/darpa_vid_search_dough/

uses...

[hesaigo999ca]

>The US military is inundated with video from airborne unmanned aircraft, remote monitoring systems and security outposts
I am wondering if they could just use a better detection system for WHEN they should start filming. A motion sensor is just a first step into limiting the amount coming in, as well, you do not need to have any video when there is nothing there, only when something is happening....with maybe a 15 second pre cache so that once motion hits, you get the 15 seconds before that event... most cameras have that, but i don't know how much of this is common in DARPA world...

And when having a acceptable level of accuracy...

[hitmark]

it will be handed over to the MPAA/NSA media "protection" branch, to help them perform their internet "police actions" against those filthy movie pirates sailing the digital seas. I expect a certain vice president to declare war on media piracy soon, as is agreed on in the election funding "agreement" between the current US president and MPAA.

Re:Slashdot reaches new degree of separation from

[Myopic]

Did I get a +5 for making a mistake and posting to the wrong article? Jeez, we often complain about moderation around here, but usually bad mods mod me down, not up. Mod me -1, Not Paying Attention!

Thanks anyway mods!