Led by Dr. Shivam Bhasin, the research team "trained" the system by having three people randomly enter 70 four-digit PIN codes on Android smartphones. A special application is installed on each mobile phone to collect data from accelerometers, gyroscopes, magnetometers, proximity sensors, barometers and ambient light sensors.
Then, the researchers use deep learning algorithms to analyze the data and match specific sensor readings with specific numbers on the on-screen keyboard. Bassin explained: "When you enter the PIN code with your mobile phone, when you press 1, 5 or 9, the mobile phone will move in a very different way. Similarly, pressing 1 with your right thumb will block more light than pressing 9. "
When the system guesses the four-digit PIN according to the response of the sensor, the accuracy rate is 99.5% (no more than 3 attempts) when unlocking the mobile phone with one of the 50 most commonly used PIN.
Basin believes that it is conceivable that people may unconsciously use technology to download malicious software to their mobile phones. After accessing the sensor of the mobile phone and obtaining the user's PIN, the program will transmit the information to the person who can unlock the mobile phone.
To prevent this from happening, he suggested that the mobile phone operating system restrict access to the mobile phone sensor so that users can only grant permissions to trusted applications.