However, the systems used by law enforcement agencies do not always have the latest algorithms. The Pinellas, for example, was last revised in 2014, although the district has evaluated other, newer products. Idemia declined to comment.
The quality gains of the best facial recognition technology in recent years have been amazing. In state tests, facial recognition algorithms compared photos to a database of 1.6 million mug shots. In 2010, the error rate under ideal conditions was just under 8 percent – good lighting and high-resolution, forward-looking photos. In 2018, it was 0.3 percent. However, law enforcement officials could not count on this reliability in surveillance situations.
Perhaps the biggest controversy in face recognition was the uneven performance among people of different races. The results of the government tests published in December show that the type of facial recognition used in police investigations leads to false-positive results when evaluating images of black women. Florida police officers said the technology's performance is not a sign that it harbors racial prejudice.
Officials in Pinellas and elsewhere also stressed the role of human review. However, tests using passport photos have shown that human inspectors also have trouble identifying the right person in a list of similar-looking facial recognition results. In these experiments, passport system employees made incorrect choices about half the time.
Poor quality images are known to contribute to mismatches, and lighting, slanted faces, or minimal panels such as baseball caps or sunglasses can affect accuracy.
In China, law enforcement agencies are trying to work around this problem by installing disruptive, high-resolution cameras with bright face-level light and linking facial recognition systems to other technologies that scan cell phones in a specific area. If a face and a phone are recognized in the same place, the system becomes more secure in a game, according to a Times study.
In countries with stricter civil liberties laws, face deficiencies have proven to be problematic, especially in systems designed to track down criminals in a crowd. A study of such a program in London, which has an extensive network of CCTV cameras, found that of the 42 matches that were suggested during testing, only eight were demonstrably correct.
Current and former officials from Pinellas County said they weren't surprised. "If you get into bank robberies and robberies in supermarkets, no – no, it doesn't work that well," said Jim Main, who handled the technical aspects of the facial recognition program for the sheriff's office until he retired in 2014. “You can don't ask how: & # 39; please stop for a second. Let me take your picture "
Kitty Bennett contributed research.