New Jersey Transit Won’t Explain How Its 'Face Mask Detection' AI Works

In 2016, the New Jersey Transit Corporation quietly ended its short-lived attempt to record and store passengers’ conversations on trains and buses after a backlash from riders and civil liberties groups. Now, the nation’s third-largest provider of bus, rail, and light-rail transportation, is quietly introducing a new form of surveillance: face mask detection.

In late January, the Federal Transit Administration (FTA) awarded NJ Transit a $600,000 grant to install face mask-detecting artificial intelligence and heat mapping systems on the River Line, which runs from Trenton to Camden. The purpose of the technology, the agencies say, is to provide customers more information through the NJ Transit smartphone app about how crowded train cars are during the time of COVID-19.

In a press release announcing the grant, NJ Transit CEO Kevin Corbett said the data collected by the systems is “expected to provide longer-term operational benefits and customer experience improvements,” but neither he nor a spokesman elaborated on what that might mean—particularly whether the agency might at some point compare images captured against facial recognition databases. NJ Transit has not yet selected a vendor to provide the technology, agency spokesman Jim Smith wrote in an email to Motherboard, but it will be working with the Rutgers University Center for Advanced Infrastructure and Transportation to study the tools’ efficacy and develop policy around their use.

“New Jersey Transit Police Department’s mask enforcement approach is not to arrest or fine anyone, but rather to educate customers to gain compliance,” Smith said. NJ Transit police have issued at least one fine, in December, to a repeat mask offender. Smith also claimed that face mask detection is not facial recognition. The FTC, however, categorizes face detection—necessary for face mask detection—as a form of facial recognition. The primary difference between the two uses is whether or not images collected by the systems are compared against a database of face templates at any given time.

Smith did not respond to a question asking whether NJ Transit would guarantee that images collected through its mask detecting system would never be used to identify riders.

Some NJ Transit riders see the move as another example of the agency overreaching in a way that is more likely to discourage ridership than bring mask scofflaws to justice.

“I don’t see much marginal benefit to this invasive technology. People have been very good about wearing masks on NJ Transit,” David Peter Alan, a long-time public transit advocate and recently retired chairman of the Lackawanna Commuter Coalition, told Motherboard. “Transit is going to have to reinvent itself and innovate [as a result of the pandemic], but what I don’t think is wise for them is to create an atmosphere of big brother-style surveillance where people who care about their privacy feel uncomfortable getting on a train.”

The grant comes at a time when security and surveillance companies are rapidly pivoting from other lines of work to sell tools like fever-detecting thermal cameras and face mask detection to public transit systems, schools, and airports. They’ve found willing customers, as agencies seek technological solutions that will hasten their return to normal operations.

The Paris subway system recently completed a trial of a face mask detection system. And New York’s Metropolitan Transportation Authority is reportedly considering thermal cameras and face mask detection for the nation’s largest subway system.

Facial recognition has been shown to perform less accurately on women and people with darker skin tones. And while algorithms’ abilities to recognize faces wearing masks have improved over the course of the pandemic, according to a recent study by the National Institute for Standards and Technology (NIST), they still fail to accurately recognize people wearing high-coverage masks between 5 and 40 percent of the time. The algorithms mis-identify people wearing masks at roughly the same rate as they do without masks.

NJ Transit’s choice to pilot the surveillance program on the River Line, which runs from Camden to Trenton, also raises other questions about its intent. Camden is 39 percent Black and 51 percent Hispanic; Trenton is 47 percent Black and 38 percent Hispanic.

Smith said NJ Transit decided to pilot the systems on the River Line because it doesn't currently have an automated system for passenger counting. The agency's other two light rail systems also don't have automated passenger counting systems, he acknowledged, but said the River Line was most conducive to face mask detecting cameras and thermal imaging because it doesn't have multiple branches.

“The River Line is a clever place for NJ Transit to institute an elevated level of surveillance,” Alan said. “I can see how if they elevate the surveillance level there, they might get less resistance for that.”

This article was updated with additional comment from NJ Transit