"Thermal Cameras and Machine Learning Combine to Snoop Out Passwords"

A team of researchers at the University of Glasgow published a paper describing their method, ThermoSecure, which discovers passwords and PINs. ThermoSecure involves using a combination of thermal imaging technology and Artificial Intelligence (AI) to expose passwords on input devices such as keyboards, touchpads, and touch screens. According to the researchers, during testing, ThermoSecure effectively attacked 6-symbol, 8-symbol, 12-symbol, and 16-symbol passwords with an average accuracy of 92 percent, 80 percent, 71 percent, and 55 percent, respectively. Furthermore, these results were based on relatively 'cold' evidence, and the paper adds that thermal images captured within 30 seconds provide even greater accuracy. The system requires a thermal camera, which has become significantly less expensive in recent years. According to the research paper, a useful device may only cost $150. In regard to AI, the system uses a Mask RCNN-based object detection technique that essentially maps the thermal image to keys. In three phases, variables such as keyboard localization are considered, followed by key entry and multi-press detection, and then algorithms determine the order of key presses. This article continues to discuss the research on thermal attacks against passwords. 

Tom's Hardware reports "Thermal Cameras and Machine Learning Combine to Snoop Out Passwords"