From rare and innovative, wearable technological devices at times embedded in a jacket, a bag or a piece of jewellery have become maybe not extremely popular, but definitely more common in the fashion industry. Yet there are researchers and designers out there who have been moving on, experimenting on other levels concerning fashion and technology.
Textile designer, illustrator and digital security trainer Kate Rose for example presented this August at the DefCon cybersecurity conference in Las Vegas a range of anti-surveillance garments from her Adversarial Fashion line.
The designs - for the time being the line comprises T-shirts, sweats, bomber jackets, dresses and skirts - are designed to trick surveillance cameras.
The garments available at the moment from the site are covered in prints of license plates and can confuse automatic license plate reader (ALPR) systems. The latter use networked surveillance cameras and simple image recognition to follow the car flows in the traffic and can collect thousands of plates a minute.
Digital systems can fail, though, as Rose explains in a presentation on her site: last November facial recognition systems accused for example businesswoman Dong Mingzhu, president of China's biggest air conditioning maker, of jaywalking because her face appeared in a bus advert. So Rose decided to explore the possibilities of confusing ALPR systems via fashion. Inspired by the fourth amendment in the US constitution protecting Americans from "unreasonable searches and seizures", Rose came up with a print design featuring a series of licence plates with the words of the fourth amendment.
The ALPR system reading garments made with this fabric will add the junk license plates to the database and will therefore get confused.
While this is the first time the creation of adversarial images regards car trackers, the anti-ALPR print is not a new idea: three years ago Berlin-based artist and technologist Adam Harvey came up indeed with a collaborative project with interaction studio Hyphen-Labs that resulted in the production of the Hyperface prototype textile, characterised by an abstract pattern that triggered and confused facial recognition systems, providing them with false faces that distracted computer vision algorithms.
In April this year Simen Thys, Wiebe Van Ranst and Toon Goedemé, three researchers at the University of KU Leuven, Belgium, also proved that it was possible to confuse an AI system. The trick in their case consisted in holding against their body an adversarial patch (AI fooling images are known as adversarial patches) representing an image of people holding colorful umbrellas that had been altered by rotating it and adding noise.
An AI system must be trained to recognise and identify things, going through thousands of objects in a specific scene. Yet, when a system is presented with something it can not recognise, it gets confused; the researchers proved that, if one of them held the patch over their mid-section, the AI system was no longer able to detect their presence.
While there are privacy issues at stake when it comes to surveillance systems, the latter are first and foremost devices to reinforce safety, spotting for example people that may be causing trouble or preventing people from driving recklessly (license plate readers may be used in tracking cases, but they are mainly employed for enforcement purposes). Researchers coming up with projects about fooling surveillance cameras are therefore not trying to provide us with methods to hack things better, but they are uncovering a vulnerability in a system and implicitly highlighting they should be changed and improved. Rose's Adversarial Fashion is indeed designed to remind us that, if a car tracker can be fooled, it can become less effective and therefore it loses its purpose.
If you're interested in exploring the possibilities of adversarial images in fashion, you should check out Kate Rose's resource library and inspiring tutorial slides from her DefCon 27 Crypto & Privacy Village Talk that can help you experimenting and making your own computer vision-triggering fashion and fabric designs.
Comments
You can follow this conversation by subscribing to the comment feed for this post.