Real-Time Detection of Transport Modes and Movement States via Smartphone Data

Accurate real-time tracking of public transport is crucial for improving passenger experience, optimizing transit operations, and enabling smart city initiatives. However, conventional public transport tracking systems primarily depend on global navigation satellite system (GNSS), which often struggle with signal disruptions in dense urban areas due to obstructions from tall buildings and tunnels. To overcome these limitations, our research proposes a machine learning framework that analyzes magnetometer data from passengers’ smartphones to detect transport modes and determine whether the passengers’ are inside a transport mode or not and also whether the transport mode is moving or stationary. This GNSS-independent approach aims to provide real-time status updates, enhancing service predictability and operational efficiency. We collected approximately 16 hours of sensor data from subways and trains in Munich using a custom mobile application. Our neural network model achieved an accuracy rate of 95% in classifying transport modes and their states and an accuracy of 98% when using an averaging filter.