The Reg has a story up about an article [PDF] published Thursday, in which researchers show off a 15 inch (40cm) square placard with a pattern that shows a high degree of effectiveness in confounding automated person-detection by surveillance cameras.
"The idea behind this work is to be able to circumvent security systems that use a person detector to generate an alarm when a person enters the view of a camera," explained Wiebe Van Ranst, a PhD researcher at KU Leuven, in an email to The Register. "Our idea is to generate an occlusion pattern that can be worn by a possible intruder to conceal the intruder from for the detector."
The technique is demonstrated in this video with the album cover sized placard centered on the body being passed between two people who disappear from recognition as it is centered on their bodies.
Looking ahead, the researchers hope to generalize their work to other neural network architectures like Faster R-CNN. They believe that they will be able turn their pattern into a T-shirt print that will make people "virtually invisible" to object-detection algorithms in automatic surveillance cameras.
Clothing such as they describe could be combined with related efforts like CV Dazzle fashion styling which is designed to confuse facial recognition software, making privacy in the future very strange looking indeed.
(Score: 2) by stretch611 on Sunday April 21 2019, @09:05AM (3 children)
It appears as if the pattern needs to be directly in front of the person and facing the camera. This means it is useless in any circumstance that you do not know where the camera is or if there are multiple cameras with different angles trying to recognize the person. Unless they can create a pattern that stops the recognition from multiple angles, I doubt this would ever be effective.
In addition, the cameras in question seem to be looking for a full size person; the pattern is at waist level. While it may take more processing horsepower, what happens when the cameras are trained to trigger on any and all faces regardless of a body being sensed? I can't see this being that hard to do as modern cameras are able to find faces quite easily... after all, that is how they determine auto-focus apertures. Again, I doubt this method would be very successful in different situations outside the specific lab test displayed in the video.
Now with 5 covid vaccine shots/boosters altering my DNA :P
(Score: 2) by c0lo on Sunday April 21 2019, @09:35AM (2 children)
The pattern seems to be an image with people and umbrelas. Go to farmer's market, take a panoramic picture, print it all around the bottom of your t-shirt.
https://www.youtube.com/watch?v=aoFiw2jMy-0 https://soylentnews.org/~MichaelDavidCrawford
(Score: 0) by Anonymous Coward on Sunday April 21 2019, @12:19PM
Do you do cargo-cult science much?
(Score: 1) by RandomFactor on Sunday April 21 2019, @01:38PM
I noticed that they kept it centered over the waistline, completely eliminating the waist separation. It is possible that the typical separation of a person at the waistline is also learned by object recognition, so breaking that up helped also.
В «Правде» нет известий, в «Известиях» нет правды