In Oregon, the Washington County Sheriff’s Department can review 300,000 mugshots in seconds and is helping other agencies to query their databases as well.
Chris Adzima, a senior information systems analyst with the Washington County Sheriff’s Office in Oregon, was already toying with the idea of using facial recognition to identify unknown suspects caught on surveillance camera when Amazon Web Services unveiled Rekognition at re:Invent last November.
Within a few days, Adzima moved 300,000 mugshots from the law enforcement agency’s database into the AWS cloud for seamless learning-based image analysis.
Rekognition is an AI tool that detects objects and faces within pictures and searches the latter against faces in a database within four to five seconds, returning up to five matches ranked by accuracy. The similarity must be greater than 50 percent, and the user can then decide whether the match is worth investigating.
“Every time we had a suspect, I started running it against our collection,” Adzima told Route Fifty by phone. “We are working with other counties to get them onto Rekognition and have their own platforms, so we can query against their collections.”
Every search run tests the validity of recognition, with Adzima keeping track of the results to gauge its effectiveness in quick apprehension of suspects—the ultimate goal. So far there have been several arrests and convictions, but not enough yet to be statistically significant.
One came early, when a theft suspect pretended to self-checkout items at a big-box hardware store before simply walking out with them, not realizing the stations come equipped with cameras that point at the customer’s face. The responding deputy sent Adzima an email with the suspect’s image, which generated a good result because of a unique upturn in the corner of the person of interest’s mouth.
Once the deputy had a name, a quick Facebook search returned a profile picture, where the suspect was wearing the same hoodie worn during the crime.
Outside of onboarding neighboring counties, Adzima is interested in using Rekognition as part of the Help Me Home program, which allows people to register the pictures and information of loved ones at risk of getting lost: seniors with dementia or children with autism. Responding deputies would be informed of how to handle a missing person they encounter after matching their photo with a smartphone, whereas now they solely rely on written descriptions.
The Sheriff’s Office has no intention of adding DMV records and photos into the Rekognition system, as there’s no indication those individuals have committed crimes in the past, and the agency is cautious of maintaining people’s privacy.
Adzima had the system fully operational within two months, at minimal cost.
“If [officials] are at all interested, just dive right in because it’s not difficult, it’s not expensive and it’s something they can do very quickly to tell if it’s going to benefit their agency,” Adzima said. “Even if only three to five people are using it, it’ll benefit them.”