Slashdot Mirror
Boston Police Chief: Facial Recognition Tech Didn't Help Find Bombing Suspects
SternisheFan writes "ArsTechnica reports: 'While the whole country is relieved that this past week's Boston Marathon bombing ordeal and subsequent lockdown of the city is finally over, Boston Police Commissioner Edward Davis told the Washington Post that the department's facial recognition system "did not identify" the two bombing suspects. "The technology came up empty even though both Tsarnaevs' images exist in official databases: Dzhokhar had a Massachusetts driver's license; the brothers had legally immigrated; and Tamerlan had been the subject of some FBI investigation," the Post reported on Saturday. Facial recognition systems can have limited utility when a grainy, low-resolution image captured at a distance from a cellphone camera or surveillance video is compared with a known, high-quality image. Meanwhile, the FBI is expected to release a large-scale facial recognition apparatus "next year for members of the Western Identification Network, a consortium of police agencies in California and eight other Western states," according to the San Jose Mercury News. Still, video surveillance did prove extremely useful in pinpointing the suspects.'"
Rinse and repeat
Computers are not magic.
Can anyone explain the presence of Craft International (Security Contractors) at the marathon?
Do Security Contractors frequently monitor events like this?
Maybe they were hired by the Boston Marathon. Or maybe your tinfoil hat is loose.
If I'm not mistaken, the CCTV footage was not as useful... what did help was the one man who took a picture of the bomber (unbeknownst to him at the time), and more importantly, the unfortunate man whose legs were blown off at the knees who valiantly gave an ID from his hospital bed.
libertarianswag.com
Gotta wonder if it picked up matches for random people who are wanted for one thing or another, and if there will be follow-up investigations on those leads.
And if so, if crowd-scanning will become a precedent...
"Boston Police Chief: Facial Recognition Tech Didn't Help Find Bombing Suspects"
Thousands of paramilitary, guns, Humvees, helicopters, robots, hours and hours of lockdown of millions of people and the suspect went uncaught.
A homeowner on a smoke break finds him.
Who the fuck cares about facial recognition, I say arm the citizens and save money and time.
The MIT policeman was apparently shot in the head while sitting in his car, not apprehending the suspects.
If Slashdot were chemistry it would look like this:Cadaverine
Yeah, right. Just ask this guy about it.
Current-generation face-recognition systems have a false positive rate of about 1 in 1,000 even when they have excellent images to work with -- high-resolution, well-lit, full-face frontal photos with no obscuring hats, glasses, etc. So even if CCTVs captured excellent images, if you're searching a database of tens of millions you're going to get a lot of matches. In a case like the Boston bombing it's okay if you get a few thousand hits because there is manpower available to sort through and narrow those down to the dozens which the (much more accurate) human eye/brain can't distinguish, and then there's manpower available to chase down each of those leads.
When you reduce the image quality, though, make it grainy, at an angle, poorly lit, and throw in some baseball caps... forget it. You have to reduce the match threshold, and then instead of thousands of candidate matches, you have tens or hundreds of thousands. For that matter, consider the fact that humans can't deal well with those constraints, and we're social animals who devote a significant portion of our enormous brain capacity to exactly this task.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
I have loaded some 45000 pictures, almost all family pics, on to Picasa. Once I identify a face and tag it, it finds the same face in other photos. And as I mark yes/no for its findings, it improves remarkably. It is not confused by heavy make up worn by Bharatnatyam dancers. It is finding the correct faces of 20 such dancers lined up facing the camera. It picks faces obscured in dark backgrounds, in out of focus pictures, faces occupying hardly 50 x 50 pixels. Faces at all orientations, including upside down. Half faces, faces with just one eye... It is really amazing.
What is amazing is its mistakes. It mistakes mother for daughter and vice versa. Confuses brothers with sisters when they are toddlers but not when they are teens or adults.
But this is forward match, going from a known face and looking for it in a crowd. Boston police is trying the reverse look up on a massive scale. It failed today. But like Lycos and webcrawler being upstaged when Google solved the reverse look up problem, some day the reverse look up problem will be solved. With parallel technology? Through GPU's running million forward searches simultaneously? But someday soon, the reverse look up will be solved and the automatic photo identification will work.
sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
So if the system doesn't actually catch bad guys, why do they still have it? Did they not save their sales receipt from spending all those tax dollars?
Join the Slashcott! Feb 10 thru Feb 17!
What people don't understand is that for facial recognition software to work you have to have good quality cameras, images, a more static environment. This is why you hear about it being used in casinos is Las Vegas and elsewhere. In those environments you have high quality cameras with close range and good angles working against a smaller set of good pictures in a relatively static environment (people in casinos tend to congregate and not move around a lot). You also have staff with a distinct vested interest in watching out for their 'bad guys'.
In a place like a large public venue you have lower quality cameras, far more people running around, worse angles and range and the environment is far more transient. The tool is being used in a completely different environment with far less support and far larger data sets to work with.
It's like taking your Rav4 off-roading the Rubicon trail and coming way with the conclusion that off-roading is a bunch of hype. You've taken the tool (grocery getter) and put it to use for a job it was never meant for. Meanwhile your guy with the old Jeep knows for a fact that his tools works for the job because he uses it for that job on a routine basis, however he would be just as foolish to except his jeep to work as well as a daily grocery getter as a Rav4.
Until the tools are put into environments that allow them to succeed, and with the hardware that they need they will continue to fail. You could call it a failing of the tool, however the tools and hardware are immature. Give it another five years and this would be a very different story. It's just technology advancing and there isn't a damn thing you can do about it short of getting hold of your politician and demanding reforms or limits on it's use.
I dunno, I ran Picasa on my pr0n dir, and it gets all the Czech models mixed up all the time.
There's your problem- you're using face recognition. Looks like there's a need for an extension so the recognition software can look further south...
> Horsehoes don't really exist.
Sure they do. They just charge more by the hour and it's tough to get a hotel room that will let them in.
Bark less. Wag more.