How can the AI ​​see us behind the walls?

Recently, a project combining radio waves and artificial intelligence (AI) has provided researchers with an interesting application, which is to track the bone structure of people moving in the way a wall.Recently, many people believe that this application was created to support the police while conducting raids, however, in practice it has been effectively applied in the medical field - to track the movements of Parkinson's patients right in their own homes.

According to Dina Katabi, a senior project researcher and professor of electrical engineering and computer science at MIT, this type of technology has received interest in the scientific community for decades. come here."There was a big project of DARPA, using wireless signals to identify people behind walls," said Dina Katabi.

However, today's technology with greater accuracy allows scientists to do more: This technology can accurately describe the bone structure of the people being followed, as well as the time they spent Perform normal movements, such as walking or sitting down.

This technology focuses on tracking key points of the body, including joints like elbows, hips and feet.So when the person being followed moves or performs movements, images that accurately describe those actions will also be displayed.

How does this technology work?

Radio signals used in this technology are almost similar to Wi-Fi waves, but are much weaker than Wi-Fi.Basically, this system works because these radio waves are able to pass through objects such as a wall.After that, the waves reach the human body but immediately they will be bounced back to the wall and go to the transceiver because the human body contains a lot of substances that prevent the penetration of radio waves, that is water.And now the problem is how to express the signals that the waves receive?And this is where the power of AI works, specifically here is a machine learning tool called the artificial intelligence network.

By allowing the AI ​​to self-study specific annotated information, artificial intelligence researchers train an artificial intelligence network that can deduce its own rules from the data learned.This process is called supervised learning.For example, if you want to teach a self-driving car to identify traffic light signals, let AI learn and familiarize yourself with traffic signal-related images with annotations so that AI can understand. Where did these images appear from.

This artificial intelligence network is often used to interpret image data, but they can also be used to perform other complex tasks such as translation or even creating new documents. by imitating the data learned.

But now, there is a problem that this technology faces, which is very difficult to handle and identify radio signals that respond to information about human body structure.And the scientists' solution is to combine radio waves with the camera.First the camera will capture the images behind the wall, then annotate these images so that the artificial intelligence network can learn the images that correlate with the activity obtained from the radio signal.

After the training period, the scientists were surprised to find that although the system was only trained with images that were not obscured, it was still possible to detect those who were hidden behind the picture. wall."He can see and simulate the shape of the person behind the wall even though he didn't learn it during the training," said Dina Katabi.

Not only that, it can even identify different people through their gait.With the help of another artificial intelligence network, the system can see examples of people walking, and from there, it can identify these people through their walks even when They are moving behind the walls, with accuracy of over 83%.

How will this technology be used?

Researchers have begun using this system in a small study with Parkinson's patients.By placing these devices in a patient's home, researchers can track patient movements in a comfortable space without using the camera.This is also a way to learn about someone's body movement without affecting their privacy as much as traditional video recording.This study applied over seven people and lasted for eight weeks.

Not only were the results consistent with the criteria used to evaluate patients, but they also revealed more information about the quality of life of Parkinson's patients, such as behavior and state of they.The Michael J. Fox Foundation is continuing to fund this study to help patients avoid the "white coat syndrom" syndrome.This is the phenomenon of patients feeling nervous, frightened and have different behaviors when seeing doctors wearing white robes for themselves.

These studies reduce privacy issues, but does not mean they will be used without the patient's consent, Ms. Katabi said.

See more:

• 6 steps to start learning artificial intelligence programming (AI)
• Entertainment on Neural Networks, Artificial Intelligence and Machine Learning
• Google's artificial intelligence "gives birth to itself", potentially far beyond human-written software
• Artificial intelligence can help detect melanoma skin cancer