Ultrasonic-Based Transportation Mode Detection in Urban Environments

Kurzfassung

Transportation Mode Detection (TMD) plays a key role in enabling intelligent transportation systems, optimizing mobility services, and supporting energy-efficient urban planning. However, methods that utilize Global Navigation Satellite System (GNSS) signals typically suffer from performance degradation in urban canyons and tunnels due to satellite occlusion and multipath interference, while Inertial Measurement Unit (IMU)-based methods are prone to cumulative errors from sensor drift. Hence, we propose in this paper, a novel ultrasonic sensing framework for classifying urban transportation modes based on vehicle-borne acoustic emissions in the ultrasonic range. To the best of the authors’ knowledge, this is the first study to analyze ultrasonic sound for transport mode identification. Using a Pettersson u384 microphone sampling at 384 kHz, we captured audio within the 20–80 kHz frequency band across four representative modes, bus, streetcar, subway, and suburban railway, during several hours of real-world operation in Munich. Each transportation mode exhibits a distinct acoustic signature that can be leveraged to differentiate between modes. We employ a stacked ensemble learning approach combining Random Forest (RF), Support Vector Machines (SVM), Logistic Regression (LR), and Multi-Layer Perceptron (MLP), with a LR meta-learner integrating the predictions. Our method achieves an accuracy of 99.57% in distinguishing between modes. This ultrasonicbased approach provides a robust, privacy-preserving alternative that complements traditional sensing modalities in context-aware mobility systems.