United States Customs and Border Protection (CBP) has initiated a request for proposals aimed at developing a real-time facial recognition system. This tool would be capable of capturing images of all individuals in a vehicle at border crossings, including passengers in rear seats, and matching these images against their travel documents, according to a recent posting in the federal register.
The request outlines that CBP currently employs a facial recognition solution that photographs individuals at ports of entry, comparing these images to identity or travel documents presented to border officers. Additionally, it uses existing photographs from documents stored within government databases. As stated in the document, entries confirmed through biometric data are integrated into travelers’ crossing records.
Operating under the Department of Homeland Security, CBP confirms its facial recognition technology is functional across air, sea, and land pedestrian environments. The agency’s next objective is to apply this system to the land vehicle context. According to information updated on CBP’s website, the agency is currently conducting tests to enhance the system’s capabilities, with the existing technology reportedly facing challenges in capturing images of every vehicle occupant, particularly those seated in the second or third row.
CBP highlights that unique factors such as human behavior and multiple rows of seats, along with environmental conditions, complicate capturing accurate images in vehicles. Consequently, the agency is seeking to collaborate with a private vendor on a solution that improves image capture of all passengers present in vehicles.
Researcher Dave Maass from the Electronic Frontier Foundation has released insights from a document obtained via a public records request, detailing outcomes from a 152-day evaluation of CBP’s facial recognition system at a port of entry from late 2021 to early 2022. The findings indicated that cameras at the Anzalduas border crossing, connecting Mexico and McAllen, Texas, successfully captured images of all vehicle occupants only 76% of the time. Furthermore, of those captured, just 81% met the necessary criteria for matching their faces with identification documents.
The system presently utilizes one-to-one facial recognition, which aligns individuals’ images with their travel documents. A key vulnerability noted is the potential for misidentification, where the system fails to recognize a person as matching their own documents. This contrasts with one-to-many recognition methods often employed in law enforcement to identify suspects, where the primary risk involves false positive identifications.
Maass has pointed out uncertainties surrounding whether the high error rates stem from the quality of the cameras or the efficacy of the matching software. There is also ongoing concern regarding potential biases linked to race and gender that could arise from these technologies.
In August, the Department of Homeland Security’s Science and Technology Directorate had previously issued a similar request for information, although that document is currently inaccessible. As CBP continues to explore advancements in facial recognition capabilities, the implications for privacy and accuracy warrant close scrutiny from security experts and business leaders alike.
