Tracking train passengers using accelerometer data from their smartphones without the use of training data

Kurzfassung

Today's smartphones are equipped with sensors, some of which can be accessed by any application at runtime without the user's permission. Attackers can exploit this to track the user's train journeys and create movement profiles. In this thesis, a framework is developed to enable the detection of train journey paths using sensor data from the victim's smartphone. Special focus is put on the accelerometer data, as it is the most suitable for this task. Training data for each train route to be recognized is not needed, unlike machine learning approaches. The detection works solely on the basis of freely available data, map data from OpenStreetMap and timetable data. From these, a graph is generated that represents the rail and train line network. This is possible for an arbitrary area. Train related data is extracted from the sensor data and matched with the graph. In the proposed framework, paths are searched based on the travel time and the start and end time. For the found paths, further features are matched with the sensor data. These are based on stop detection, driving directions and timetable data. Each path is given a score, which indicates how likely it is that the path was actually driven. It is also possible to consider journeys consisting of consecutive sections between which there can be any changeover times. In the evaluation it could be shown that the correct path always belongs at least to the 16% of paths with the highest scores. In the final test case, the actual path was even the one with the highest score. It was also found that the correct functionality of the stop detection and the presence of train delay data can greatly improve the results.