Inferring Touch from Motion in Real World Data

Most modern smartphones are equipped with motion sensors to measure the movement and orientation of the device. On Android and iOS, accessing the motion sensors does not require any special permissions. On the other hand, touch input is only available to the application currently in the foreground because it may reveal sensitive information such as passwords. In this paper, we present a side channel attack on touch input by analyzing motion sensor readings. Our data set contains more than a million gestures from 1'493 users with 615 distinct device models. To infer touch from motion inputs, we use a classifier based on the Dynamic Time Warping algorithm. The evaluation shows that our method performs significantly better than random guessing in real world usage scenarios.