I wasn’t able to jump on this yesterday due to some Military.com commitments, but there was an interesting piece in the New York Times about the huge amount of UAV data pouring into military hard drives — so much that the USAF, for example, is drowning in it.
It got me to thinking that the services are exactly right to store all that drone feed footage no matter how boring it might be. The flight of a Reaper drone from its impoundment in Jalalabad to its target in Miran Shah might be just rocky paths and scrub brush, but to a skilled analyst, the tell-tale differences from each pass over a span of time might mean the difference between detecting a new “rat line” and ignoring a key Taliban infiltration route.
A group of young analysts already watches every second of the footage live as it is streamed to Langley Air Force Base here and to other intelligence centers, and they quickly pass warnings about insurgents and roadside bombs to troops in the field.
But military officials also see much potential in using the archives of video collected by the drones for later analysis, like searching for patterns of insurgent activity over time. To date, only a small fraction of the stored video has been retrieved for such intelligence purposes.
The story seems to indicate that there’s a shortage of analysts to evaluate the video and pinpoint the intel that might prove useful — especially if it’s second or third order data.
Air Force officials, who take the lead in analyzing the video from Iraq and Afghanistan, say they have managed to keep up with the most urgent assignments. And it was clear, on a visit to the analysis center in an old hangar here, that they were often able to correlate the video data with clues in still images and intercepted phone conversations to build a fuller picture of the biggest threats.
But aren’t there software solutions that can process the footage and pick out the things analysts might be interested in? I mean, the National Geospacial Intelligence Agency doesn’t pour over hard copies of Key Hole satellite shotsKestrelTest with a magnifying glass anymore, do they?
But while the biggest timesaver would be to automatically scan the video for trucks and armed men, that software is not yet reliable. And the military has run into the same problem that the broadcast industry has in trying to pick out football players swarming on a tackle.
So I dredged up a company I’d seen one year at a trade show that developed software to run in the background of UAV feeds. The application pinpoints vehicles, personnel and other objects interesting to the operator and tracks them in a color coded box. Seems to me the same could be developed for a passive application where the video footage is just run through the processor after the mission and the software picks out certain clips that contain the clues analysts program in.
I can see the article’s point — the AF is developing new software to get key info to the field from drone passes faster to the operator on the ground…but what about that change analysis piece?
Here’s a cool analysis test from the Kestrel web site.
– Christian
{ 12 comments… read them below or add one }
Hire mor anaylists, use new 3D visual computer software
Improve images alone.
Use dbase of say missiles, rockets etc from Cold War for Reference Dbase.
Use 3D Terrain mapping software.
More can be done.
Outsource to Pvt Security Companies like Xe?
shut the fuck up, everyone who knows a sliver of the intel. world thinks contracters are more qualified than the mil. theres a big difference, the contractors sell better because they can, but the individual contractor on the floor sucks. they are usually some retired vet. who is in it for the paycheck. they are fucking lazy
Mechanical Turk is the answer.
But dear what shall they wear ?
A super computer should be created for each war theatre to absorb all of this information & any documented enemy altercations.
Would think these were already in place.
As well as a satellite to monitor all moving vehicles as much as could. The key feeds would show truck bombs route & origins.
Movies tend to show america with some vast secret underground control center with tech like this. I think it's time to start to build it.
As to me , A civilian observer of this 'war'sees only that this is a war against crime. Terror is just a violent & deadly crime. No country in history has defeated crime.
if we are to use tech to defeat the enemy. We must step it up now, while we still can.
I have a friend who is working on the problem of image analysis and it is not an easy problem. When you are talking about telling a computer to sift through relatively low resolution imagery it is easy to have it flag the t-72 or even to measure traffic patterns but but it is much less easy to talk about differentiating between some kids playing soccer and some insurgents burying IEDs or to identify what would constitute anomalous traffic behavior. There are two ways that you can approach the problem; you can decrease the sensitivity of the program (i.e., have the software flag every group of people for further inspection) or increase the precision (i.e., get it to properly distinguish between the kids and the terrorists.) The former solution is trivial but almost useless due to the quantity of data floating around and the latter is anything but trivial. It might be very clear who planted the bombs as you are backtracking through surveillance leading up to an explosion, but this is only because you have such an obvious identifying marker as a big orange and black thing.
flagging using human monitors at time of feed is all is needed. The computer would only be used to process history & the different sources into essientially folders. Traffic, our movements, identified enemy killed captured or encountered info also sent with timestamp & location
The application then would only need to reference history & infodata to be displayed in a preset map of the battlefield.
From there 3D CGI Virtualize it all you want, but the basics are easily capable. Human analysts & tactical commanders could pull up locations, time periods & see statistics with known tactics used against us…
I have designed and implemented imaging algorithms that detect, track, and classify targets of interest. It is not so hard to accurately find people, vehicles, and other objects. But as Nathan points out figuring out who the "good guys" and "bad guys" are is where things get really tricky. ____I don't think intelligence analysts will be replaced anytime soon. People are very well trained classifiers- we can pick up on very subtle clues that are very difficult to get a computer to recognize accurately. Nothing beats a man in the loop for identifying threats, especially with low resolution imagery that military surveillance cameras typically have. ____I think the solution is what Rob says- use software to flag suspicious activity but keep a man in the loop for final inspection and verification. This alone is a large reduction in human involvement. As software incrementally improves the need for a man in the loop will diminish.
You sum it up perfectly. Although, I would love to see more effort to map any possible patterns of enemy troop movement, etc.
I read UAV Data Hose & most of subsequent responses with great interest.From 9+years of historical NASA, NSA Keyhole, Satellites, U-2, Drone overflight data recorded,via EDI,Infrared, Microwave Ground Penetrating Radar, of the distinct typology of Iraq, Iran, Afghanistan, W Pakistan, I think we can creat a 3-D, typographical, geographical digital overlay, sterographic 3-d map Baseline with GPs indicators of underground C&C Bunkers, Weapsons Chaches, Underground Tunnels, Foot Trails, Rat Lines. Moving forward, we can measure daily changes tp these sites by digital variance analysis comparison back to the digitial baseline created from multiple souirces listed.Box Jenkins statistics uses 7 years of statistics for 95%+ level of confidence. In light of $30.0 M-$60.0 M per day for the wars lease as many Cray Super Computers as required to store, parse, analize this massive incoming data stream. Exception processing to ID exceptions to static aspects of base line data created for variance analyis measurements.Cost is not an issue in the overall scheme of Risk Mgt. We need to integrate these disparate sources of raw UAV Data for Target ID, Target Acquisition, Target Distruction.
"But aren’t there software solutions that can process the footage and pick out the things analysts might be interested in? I mean, the NGIC doesn’t pour over hard copies of Key Hole satellite shots with a magnifying glass anymore, do they?"
BWAAAHHHAaaahhaaahhaaaa….
Ok, let me clean up the coffee that just came shooting out my nose.
As a contractor and soldier (title 32) you just made my day Christian.
I wish we had 1/10th of what you guys seem to think we have with regard to connectivity and "super computing"…
Real life is not like what you see the actors on "24", "Alias", "The Unit", "CSI Miami" or "NCIS".
I freaking wish.
War would have been over in a month… we could have just hacked into the Iridum and Thuryia networks and manually caused all of the batteries in Al Q's sat phones to short out, over heat and explode when ever our petafolp crunching Deep Blues voice recognition software sensed a HVT talking on the line.
Handset goes BOOM. Dead in your cave.
You guys have no idea how jacked up and disconnected shit really is.
~JOS
The problem is that our technology is overtaking our ability to deal with the flood of information. We see the effects of Information Overload in the business world every day when people miss critical e-mails or important information.
In the case of drones (and other security intelligence), the consequences of missing key information are life-and-death, and a lot more serious than a few missed e-mails.
I don't think technological solutions are the answer, we first need a better management science for people that understands of how to deal with the flood on information that these tools are sending back to us.
I wrote a short piece on this topic, any thoughts are welcome.
http://www.basexblog.com/2010/01/14/intelligence-...
Thanks.