Ugly Sweater tricks security cameras and artificial intelligence systems

Those who take a critical look at the growing surveillance of public spaces can apparently defend themselves more easily than they thought. Researchers at the University of Maryland have developed an “ugly sweater” that misleads security cameras or the AI ​​behind them to assess recorded material.

The underlying scheme works quite simply: algorithms recognize people and faces based on their training, allowing them to distinguish certain patterns in the video image. However, because the sweater is printed to simulate a normal environment and uses special contrasts and patterns, the AI ​​becomes confused and in many cases cannot correctly identify the person wearing the garment.

However, the whole is not as reliable as one might wish: the AI ​​can then distinguish 50% of a person in the end. It probably depends on various factors such as height, weight, hairstyle, skin color, etc. Employees of Facebook Artificial Intelligence also worked on the project.

Why the whole project? Originally, we wanted to detect errors in the algorithms in order to possibly correct them. This is how the potential of the ugly sweater was discovered. Some readers would certainly like it under the Christmas tree.

This article contains affiliate links, so we mark it as advertisement. By clicking on it, you will directly access the supplier. If you decide to make a purchase there, we will receive a small commission. Nothing changes in the price for you. Thank you for your support!