The Department of Homeland Security has started a program that will test the accuracy of various off-the-shelf facial recognition platforms.
Facial recognition has come a long way in the last few years, with personal handheld digital cameras now offering basic biometric applications. But on a grander scale, the U.S. Department of Homeland Security (DHS) has embarked on a program to better understand facial recognition and how it can be used by all levels of government and first responders.
The DHS is in the midst of a project where video footage taken of people moving throughout the Toyota Center in Kennewick, Wash., will be combined with mock profiles of volunteers. Then various commercial off-the-shelf facial recognition products will be tested to see how accurate they are.
The data being collected consists of video taken during home games of the Tri-City Americans of the Western Hockey League. The Pacific Northwest National Laboratory (PNNL) is handling video collection. Results from the facial recognition assessment will be made available to the government.
“What we’d like to see is how well the current facial recognition systems perform,” said Patricia Wolfhope, program manager in DHS’ Science & Technology Directorate. “The only way to do that is to compare the data we are going to get from this data collection with the facial recognition algorithms.”
Wolfhope added that the assessment was about understanding the overall state of the technology and not necessarily designed to target use in a specific setting or with a specific application in mind.
The directorate is responsible for all of the research, development, testing and evaluation of technologies and standards and other solutions to meet the challenges faced by the operational components of DHS and the law enforcement, emergency management and the first responder communities.
The DHS research arm works with emergency response agencies to find out what they need to make their jobs better, safer, faster and more efficient and then reaches out to universities, laboratories and the private sector to find out who has solutions. Sometimes it can tweak applications available to meet the needs or, if necessary, it can build solutions from scratch.
The facial recognition algorithms will be supplied by vendors who currently have commercially available technology available. The DHS is currently in the data collection phase of the project.
“What we are doing right now is creating the video data that will be used to test everything and until we have that standardized and ready to go, I don’t think we’re going to call for manufacturers to start sending us their algorithms until we know how we are going to test,” said John Verrico, chief of media relations at the directorate.
An approximate date on when that call to vendors would be made was not available.
Wolfhope said one of the primary goals in the assessment is to see how many times a person in the video is successfully picked up by a facial recognition system versus false positives. She added that the minor league hockey arena is a good venue for video collection, because it mimics similar types of areas the DHS is interested in testing against.
Verrico noted that the venue “offers a place that has a higher volume of people,” and a high throughput of people moving through checkpoints such as where tickets are collected or where fans are buying concessions. “It enables us to use a crowd scenario,” he said.
Wolfhope specified that cameras are only being used on the concourse only and are not being used in the seating area. She also said that fans were given the option to opt out of those games where data collection was occurring, but no one thus far has.
The first video was collected at a game in September, and two or three more are expected. No other venues are expected to be used as part of this assessment, which from start to finish is expected to take about a year’s time.
According to Wolfhope, none of the video footage collected will be used against existing databases. Only the profiles of the volunteers are being used to assess the strengths and weaknesses of various facial recognition algorithms.