Invisible T-Shirt Tricks Surveillance Cameras

A team of experts from Northeastern University in Boston, the Massachusetts Institute of Technology, and IBM has developed a t-shirt that makes a person invisible to surveillance cameras. According to Wired, the t-shirt looks just like any other, but it features a special pattern of abstract, multicolored spots that can confuse the artificial intelligence used in surveillance systems.

One of the project’s authors, Associate Professor of Electrical and Computer Engineering Suyeon Lin, explained that the artificial neural networks used in object recognition systems identify someone or something as an image, draw a box around it, and then classify it. This box helps the neural network determine whether something is an object or not. Lin and her colleagues managed to calculate the boundary points of this box and create a pattern that can “convince” the neural network that there is no object in front of it.

Working with two object recognition neural networks, YOLOv2 and Faster R-CNN, the researchers identified areas of the body where adding pixel noise could confuse the AI and make a person invisible to it.

Similar research has been done before, but only with static materials. For example, in 2017, American researchers managed to trick a neural network into mistaking a stop sign for a speed limit sign. However, it is much more difficult to fool a surveillance system, where the object must remain invisible throughout the entire recording. “When recognition happens in every frame, it’s very hard to stay completely undetected,” noted Battista Biggio, Associate Professor at the University of Cagliari in Italy.