Sheriff's office spins up cloud-based facial recognition
- By Sara Friedman
- Jun 20, 2017
Law enforcement officers in Washington County, Ore., are using a cloud-based facial recognition service to identify persons of interest. Rekognition, available through Amazon Web Services, leverages an application programming interface that allows police to build apps that match pictures of suspects against a pool of mugshots to identify alleged criminals with no less than 80 percent accuracy.
When Rekognition was initially announced in December 2016, officers saw it as a way to improve on the traditional approaches of releasing mugshots to the media or passing the images among detectives.
Chris Adzima, senior information systems analyst at the Washington County Sheriff’s Office, decided to look into Rekognition when he was trying to find a faster and less expensive way to narrow down the list of suspects for crimes across the county.
“I had not used AWS before for any of our services, but I was asked to look into facial recognition because we were noticing that we are getting more requests to identify people,” Adzima told GCN at the recent AWS Public Sector Summit. “We have such a large pool of mugshots to pull from, and this product was announced essentially about a week after I was asked to look into this, so it was the right timing.”
It took Adzima three days to move 300,000 mugshots collected since 2001 by the Washington County jail management system to an AWS S3 bucket. Through a PHP script, Adzima was able to index the images to make sure the program would provide the highest match for law enforcement officials. He also built a user interface to make it easier for law enforcement officials to use the service.
“The specific customization that I really liked about Rekognition and AWS as a whole was because it is all API based,” Adzima said. “I’m able to utilize the service, but make it my own and integrate it with the framework that we already have for our web server.”
Adzima finds many law enforcement officers are hesitant to learn new systems for their jobs, but he said the Rekognition-based service is attractive to them because it is “kept in the framework that we already have.”
In Washington County, Adzima said Rekognition is having some problems with the highest match incorrectly identifying a suspect.
“In fact, most of the time a hit will come in with the highest probabilities being 90 percent similar, 85 percent similar or down from there, but the hit with the highest match won’t be the priority hit,” Adzima said. But he predicted that more machine learning capabilities will be added into the platform over time. “It would be nice to have them learn from the hits and misses so we are able to get more accurate results,” he said.
Rekognition is currently the only service the sheriff's office runs on the AWS US West public cloud, with the rest of the systems operated on premise.
Washington County doesn’t submit any information governed by Criminal Justice Information System policies through Rekognition. The images submitted for facial recognition don’t have any identifiable information attached to them that isn’t publicly accessible. When suspects are booked in Oregon, information such as their names, dates of birth, height and weight released to the public.
Last week, AWS announced Rekognition is now available in the GovCloud region for public-sector customers in the United States, which will make it accessible for CJIS-compliant applications.
More details on how Adzima configured Rekognition for his police department is available in an AWS blog post.
Sara Friedman is a reporter/producer for GCN, covering cloud, cybersecurity and a wide range of other public-sector IT topics.
Before joining GCN, Friedman was a reporter for Gambling Compliance, where she covered state issues related to casinos, lotteries and fantasy sports. She has also written for Communications Daily and Washington Internet Daily on state telecom and cloud computing. Friedman is a graduate of Ithaca College, where she studied journalism, politics and international communications.
Friedman can be contacted at firstname.lastname@example.org or follow her on Twitter @SaraEFriedman.
Click here for previous articles by Friedman.