• rumba@lemmy.zip
    link
    fedilink
    English
    arrow-up
    0
    ·
    16 天前

    A decade ago we were capable of accurately measuring Interpupillary distance and distance to ears, shape of face. It was really great for a head on stuff but not so much for low quality security cameras at obtuse angles.

    AI is both frighteningly accurate and frighteningly dangerous for inaccuracy. Facial recognition is still really bad with dark skin. AI training is often underrepresented by people who are non-white. For example Google’s doorbell AI enumerates everyone who looks vaguely Korean as the first Korean person that was registered. Of the 17 women Latino that clean our business they all enumerate as one of three people. Are black 130 lb 6 ft tall delivery guy absolutely shows up as our 6-ft tall 250 lb project Manager. They look nothing alike but the lighting is just overhead fluorescent and it can’t see any features with those crappy sensors

  • antlion@lemmy.dbzer0.com
    link
    fedilink
    English
    arrow-up
    0
    ·
    16 天前

    It may also be possible to identify somebody by their gait. Be sure to practice the Ministry of Silly Walks before committing a crime.