Student-made invisibility coat aims to hide wearers from AI cameras

The accuracy of pedestrian identification was reduced by 57% when the students tested the outfit on on-campus security cameras.
Christopher McFadden
Face recognition technology
Face recognition technology


According to the South China Morning Post (SCMP), Chinese students have successfully developed a coat that can make people invisible to security cameras. So the SCMP story goes, the coat looks the same as regular camouflaged clothing, but it can trick digital cameras, especially ones with AI.

This is achieved, so it is claimed, by virtue of the patterning of the coat that was developed using a complex algorithm. The coat also comes with inbuilt thermal devices that can emit various temperatures at night.

Since many security cameras typically use infrared thermal imaging to identify people, the coat's ability to create a distinctive heat pattern is another way of avoiding detection at night.

Called the "InvisDefense" coat, it was created by a team of four graduate students from Wuhan University in China. The coat was also one of the proposals that won first place in the first "Huawei Cup," a cybersecurity innovation competition sponsored by Chinese tech giant Huawei.

Wei Hui, a Ph.D. student in computer science who created the coat's fundamental algorithm, told VICE World News that "we spent a lot of energy preparing for this, including this product’s design and development." He also told VICE that the InvisDefense skin gives you a "novel" way to get around the AI technology that security cameras use to find people.

The accuracy of pedestrian identification was reduced by 57% when the students tested the coat on on-campus security cameras. Finding a balance between deceiving the camera and human sight was claimed to be one of the coat's primary development challenges.

"We had to use an algorithm to design a least conspicuous image that could render camera vision ineffective," he told VICE.

China has a very sophisticated system of surveillance cameras

China has a well-known, high-tech system of government surveillance that is known to invade people's privacy and go after political opponents of the regime. In fact, according to VICE, eight of the world's top ten most-watched cities were located in this nation in 2019.

Today, the government and businesses use AI identification tools, from identifying "suspicious" Muslims in Xinjiang to deterring kids from playing video games late at night.

There has been some opposition, as you'd expect, but it hasn't been able to get the cameras removed. In the nation's first-ever case contesting facial recognition technology, a law professor successfully sued a zoo in Hangzhou in 2020 for gathering visitors' facial biometric data without their consent.

The Wuhan University researchers considered similar privacy issues when creating the InvisDefense coat, which costs roughly 500 yuan ($71).

“Security cameras using AI technology are everywhere. They pervade our lives,” Wei explained to VICE. “Our privacy is exposed under machine vision.” 

“We designed this product to counter malicious detection and protect people’s privacy and safety in certain circumstances,” he added.

According to Wei, the team's future research goals include making inanimate objects "invisible" to AI cameras and making cars "invisible" to AI cameras. They are also investigating ways to get around cameras that rely on remote sensing, satellites, or aircraft.

The researchers, who are Chinese citizens, are not trying to stop the government from keeping a close eye on everyone. Instead, the team claims that they want to make it stronger.

“The fact that security cameras cannot detect the InvisDefense coat means that they are flawed,” said Wei. “We are also working on this project to stimulate the development of existing machine vision technology, because we’re basically finding loopholes,” he added.

Add Interesting Engineering to your Google News feed.
Add Interesting Engineering to your Google News feed.
message circleSHOW COMMENT (1)chevron
Job Board