Artificial Intelligence surveillance is overhyped. Part 1: tips to escape facial recognition

Surveillance technologies are getting widespread and more powerful. For example, half of American adults are now recorded in police facial recognition databases. However, these technologies do not target only criminals and terrorists. They are used in many other situations. They can identify peaceful protesters in a crowd. They can monitor employees at the workplace. They can track visitors in a retail shop. Surveillance technologies are watching us as citizens, employees, and customers.

In this context, it becomes important to understand how these technologies see us, and even more important, to learn how to escape from their sight. Indeed, despite some impressive results, surveillance technologies still suffer from numerous blind spots. At Mindolia, although we sell these AI surveillance softwares, we also feel the need to publicly reveal their weaknesses.

In this series of posts, we expose some of these cracks. They are not secret. Most of scientific research is public, and widely disseminated on the internet. However, this research is not directly accessible to a large audience, because scientific literature is hard to read for non-specialists. Here, by using simple terms, we expose those cracks found in the scientific literature.

Facial recognition

Most systems of facial recognition require the face to be frontal. They only allow for small angle variations:

In practice, people do not look at the camera

However, in practice, people do not look straight at the surveillance cameras. Therefore, in many situations, the angle between the face and the camera is too wide, and frontal facial recognition fails. To tackle this challenge, some recent methods are designed to specifically recognize the person from her profile:

There are specific methods to detect the profile.

But overall, technology is still struggling with non-frontal facial recognition, and research remains very active in this field.

You got lucky if:

You get caught when the angle between your face and the camera is too wide. For example, the following situations are hard for a surveillance system, and probably for humans too:

Who is this celebrity? If you find, tell it in the comments!

Tips to escape:

Avoid looking straight at the cameras. Look down, and wear a cap, hat or scarf, to hide from ceiling cameras. To hide even more, you can try to wear a specialized equipment that simply blinds cameras, although this might attract unwanted attention from humans:

And tomorrow?

Technology keeps increasing the capacities of surveillance systems. For example, one active technology is image completion. Soon, it will be possible to reconstruct a facial composite of a person by using only fragments of the image of her face. This synthesized face might then be fed into a facial recognition system.

Do you have any tips to escape this method?

In the second part of the series, we explore some non-facial methods of person recognition, and reveal tips to hide from them.

We sell our AI surveillance softwares here: www.mindolia.com.