Inferring User Routes and Locations Using Zero-Permission Mobile Sensors

Leakage of user location and traffic patterns is a serious security threat with significant implications on privacy as reported by recent surveys and identified by the US Congress Location Privacy Protection Act of 2014. While mobile phones can restrict the explicit access to location information to applications authorized by the user, they are ill-equipped to protect against side-channel attacks. In this paper, we show that a zeropermissions Android app can infer vehicular users' location and traveled routes, with high accuracy and without the users' knowledge, using gyroscope, accelerometer, and magnetometer information. We modeled this problem as a maximum likelihood route identification on a graph. The graph is generated from the OpenStreetMap publicly available database of roads. Our route identification algorithms output both a ranked list of potential routes as well a ranked list of route-clusters. Through extensive simulations over 11 cities, we show that for most cities with probability higher than 50% it is possible to output a short list of 10 routes containing the traveled route. In real driving experiments (over 980 Km) in the cities of Boston (resp. Waltham), Massachusetts, we report a probability of 30% (resp. 60%) of inferring a list of 10 routes containing the true route.