Feasibility of Inferring Keystrokes on PEDs with Sensors from Mobile Devices

Sensors in smart devices can help people in their lives. But data collected by these sensors can be sensitive and cause information leakage. In this paper, when inputting keystrokes on an ATM keyboard, we consider data from the accelerometer and microphone in mobile phone and smart watch concurrently. We extract features from this sensor data, show the nature of this data and analyzed the possibility of using this data to distinguish keystrokes. Furthermore, we launched a simple experiment, trained a set of classifiers to infer keystrokes using the time difference of arrival features. The result shows the potential and feasibility of using this data to perform side-channel attack and infer keystrokes on ATM keyboard.