TAMPA, Fla.—The FBI is working toward creating an enterprise solution for how it uses video analytics, looking for ways to reduce manpower requirements, increase automation, make sure evidence is properly handled and ensuring it creates investigative leads while adhering to civil liberties, an agency official says.
The technology isn’t ready now but “I have no doubt in a few years” video analytics is “kind of going to be [in] day to day use [and] we’re going to be using video analytics capabilities to do some things that we really don’t think we can do right now,” Scott Swann, special assistant to the Executive Director for Science and Technology, says at the annual Global Identity Summit this month. “So we’re trying to put our hands around exactly what we want to do with video from the very beginning, from the ingest piece all the way to the back end system.”
One of the key catalysts for the FBI beginning to examine more closely how it can make use of video analytics was the April 2013 bombings at the Boston Marathon that were carried out by two brothers that were homegrown extremists. Swann says that following the bombings, which resulted in three deaths and numerous injuries, law enforcement agencies received about 126,000 images and 12,300 videos from the public, a substantial challenge in terms of managing the data.
Three days after the attack, the suspects were identified.
In Boston it “was all hands on deck” but the FBI doesn’t have a good strategy for how its examiners would manage and look at this data so it “needs a game plan,” Swann says.
The FBI is lining up a three phase plan to test video analytics, with the first phase about to begin. Swann says in Phase I, which is slated to be completed around April 2015, the demonstrations won’t be complex and will look at things like proving that the amount of video that needs to actually be reviewed can be significantly reduced because nothing is happening and that it can provide “actionable leads.”
In Phase I things like face recognition, motion and object detection will be looked at, he says.
The data will be presented to FBI leadership as part of the effort to move forward with video analytics, Swann adds.
The FBI’s general counsel will also be involved “to see what we’ll be able to put our video lens to,” Swann says.
The shaping of the second phase will be dependent on the outcome of the first, Swann says, adding that it will be more complex and will include more data sets, the handling of analytics on edge devices rather than in a central location, and piloting analytics using existing systems like the bureau’s Next Generation Identification (NGI) system biometric database.
The third phase will consist of putting together the findings and lessons learned from the first two phases and figuring out how to move forward, he says.
Video Analytics Study
A month after the bombings, the FBI began a Major Issues Study of video analytics and in Oct. 2013 it issued a Request for Information about video analytics that resulted in 133 responses from 100 vendors, Swann says. In Nov. 2013 the report was finished and two key deliverables were lessons learned to help guide what an enterprise capability would look like and in April 2014 a capabilities roadmap with strategic recommendations for near, mid and long-term needs, he says.
Some of the key findings from the study, according to Swann, are that the FBI has various groups already dealing with video and communication between them is improving, that there is a lot of video being captured, that requirements across the government such as identifying people and objects are similar despite different missions, industry already has a lot of capability if not a turnkey solution, and the research base is robust, creating opportunities for partnerships that the agency can leverage and steer to its benefit.
There are plenty of challenges going forward, Swann says. One is cost. Generally, the better the quality of the data the more that can be done with it but it comes at a price, he adds.
In a video demonstration exhibited by Swann, government-off-the-shelf video analytics coupled with a high definition camera found 24 faces versus nine for a standard definition camera. Infrastructure costs with a high definition camera can be three and a half times higher than with a standard definition camera, Swann says he’s been told, with a 1,000 camera network the cost could potentially be “prohibitive.”
This means that “smart decisions” will have to be taken about the use of analytics, such as when data is “egressed,” Swann says. He adds that the FBI also wants more automation with regard to video analytics.
“These are the kind of things we’re looking at right now and make decisions for our enterprise about how we want to do this and we don’t quite know yet,” Swann says.
The FBI’s current biometric data base, the NGI system, mainly houses fingerprint data although other modalities that are being stored and matched, and in some cases pilot tested, such as palm, face and iris. The database also has a repository for scars, marks and tattoos but this is manually searched.
Searches against facial images require an open case and a probe image that has been lawfully obtained, Jeremy Wiltz, deputy Assistant Director of the FBI’s Criminal Justice Information System, tells attendees at the conference. The agency doesn’t use NGI to go through crowd scenes to look for fugitives, he says.
Facial searches are done against a database of mug shots, typically resulting in about 50 candidates that are then manually examined and whittled down to one or two images, Wiltz says This is a lead, not “an identical match” and “we have no way to prove right now that it’s an identical match,” he adds.
Lockheed Martin [LMT] is the prime contractor for NGI, which this week achieved full operational capability.