AI License Plate Readers Are Cheaper—So Drive Carefully

The town of Rotterdam, New York, has only 45 police officers, but technology extends their reach. Each day a department computer logs the license plates of around 10,000 vehicles moving through and around town, using software plugged into a network of cameras at major intersections and commercial areas.

“Let’s say for instance you had a bank robbed,” says Jeffrey Collins, a lieutenant who supervises the department’s uniform division. “You can look back and see every car that passed.” Officers can search back in time for a specific plate, and also by color, make, and model of car.

Rotterdam’s program is an example of how improvements in artificial intelligence software provide cops, companies, and governments new or expanded powers of surveillance and investigation.

The tech industry’s current enthusiasm for AI was kindled by a research breakthrough in 2012 that vastly improved the ability of software to recognize objects in photos. One result is progress on still-nascent projects such as autonomous vehicles and software that diagnoses cancer. In the real world, more straightforward applications of the technology have made tracking faces or license plates much cheaper and more accurate.

Automated license plate readers, or ALPRs, first appeared at police departments in the 2000s, as specialized and expensive cameras. Collins says today those devices typically cost $15,000 to $20,000. But last year Rotterdam embraced a newer generation of ALPR technology, software that can discern plates from more or less any conventional security camera. Rotterdam’s supplier Rekor Systems charges as little as $50 a month to read plates from a single camera.

“The software is a lot more cost effective than buying a full system,” says Collins. “That can change everything.” Drivers in Rotterdam used to be watched by three conventional license plate readers, two fixed and one mounted to a police vehicle. Now, five of the town’s public security cameras also are connected to Rekor’s software, significantly expanding the police’s view of the movements of local vehicles.

What happened in Rotterdam is the latest version of a now-familiar story: Like microchips and smartphones, ALPR technology is flipping from exotic to ubiquitous. Because AI software helps computers make sense of the real world, the effects of democratizing the technology can be particularly striking—and to some concerning.

Rekor’s technology has also widened the net cast by police in Sands Point, New York, an affluent village on the north shore of Long Island. When police chief Thomas Ruehle discovered the company at a conference last October, his department, with a staff of just 20, already had a conventional license plate reader system bought more than a decade ago. Within months his officers began logging the plate of every vehicle going in and out of Sands Point, thanks to seven security cameras linked to Rekor’s software. “We have every access point to our village covered,” Ruehle says. “It makes it a gated community without putting a gate up.”

Ruehle says the new software is significantly more accurate than his department’s conventional ALPR system, correctly reading plates about 97 percent of the time, compared with about 80 percent for the older technology.

Rekor’s technology originated in open source software called OpenALPR released in 2015 by Matthew Hill, now the company’s chief science officer. He was working as a software engineer elsewhere at the time and wanted a side project that would help him learn more about machine vision. To his surprise, OpenALPR quickly drew notice from people with a professional interest in reading license plates. “People were comparing my project that took six weeks against these legacy ALPR cameras and the accuracy was on par,” Hill says.

Hill built a company around the project and made the technology more accurate by switching its conventional image-deciphering algorithms for the neural network technology that started the recent AI boom. His company was acquired last year by Novume, a publicly listed security company later renamed to Rekor. Hill’s technology is now used by police departments across the US, the Department of Defense, and offered by Nokia to cities trying to improve their digital infrastructure.

Rekor is not the only company using neural networks to widen access to license plate readers. Flock Safety, an Atlanta startup, is putting the technology into private citizens’ hands by selling it to neighborhood watch programs.

Axon, a leading supplier of police body cameras and dash cams, announced last fall that it would offer license-plate reading as an add-on to its in-car video system in 2020. On the same day, the company’s AI ethics board of external experts in policing, law, and technology released a 54-page report calling for new legal restrictions on use of ALPRs, which now are largely unregulated. It cited evidence that the technology disproportionately directs more police attention and enforcement onto lower income communities and people of color. Axon says it is using the report to inform the design of its ALPR service, and is still “very early in the planning stages of this project.”

Conventional ALPRs are also getting a boost from AI. Canada’s Genetec, a leading supplier of surveillance and security technology, recently announced a new ALPR camera for use on police vehicles that includes an Intel chip designed to power image-processing neural networks. Genetec says that will make the device better at reading plates at high speed or in bad weather.

Genetec has also begun to offer ALPR software that integrates with conventional security cameras, as Rekor and others do. But company vice president Pervez Siddiqui says dedicated hardware gets better results in tough conditions like rain or snow, in part because it captures better images. “In challenging environments the only reliable solution is to use a high-performance ALPR camera,” he says.

Daniel Schwarz, who works on policy at New York Civil Liberties Union, says cheaper ALPR technology makes it more urgent to place restrictions on its use. He says tracking license plates can give government agencies too much knowledge about citizens’ private lives. NYCLU has used public-record requests to reveal the spread of the technology in New York state, including the New York Police Department’s contract with Vigilant Solutions, which sells police departments access to a database of more than 6 billion US ALPR records. Rekor is launching its own database, which it says will aggregate 30 million plate readings a week from state and local law enforcement customers.

“Widespread deployment is creating invasive databases with a comprehensive record of people’s movements,” Schwarz says. “That can show whether you’re going to a certain medical clinic, your political interests, and religious beliefs.” Documents obtained by WIRED in 2018 showed Los Angeles law enforcement agencies made tens of thousands of license plate queries each year, with help from a system built for LAPD at a cost of $6 million by Palantir, a data mining company best known for working with intelligence agencies.

Some states have laws restricting use of ALPRs but most do not. California’s Highway Patrol must delete ALPR data after 60 days unless it is being used as evidence of a felony, and all California agencies using the technology must post privacy policies online. New York is among the majority of states that have no such laws, although the state legislature is considering a bill that would limit data retention to 180 days.

Asked about his department’s data retention policies, Collins of Rotterdam Police Department said it keeps license plate records “as long as we possibly can.” He and chief Ruehle of Sands Point both say their local communities have been supportive of their expanded use of the technology. “The public feedback has been positive,” Collins says. “We use it everyday as another set of eyes for us to conclude our investigations.”

More Great WIRED Stories

Read More