Proposal of Acoustic Train Detection System for Crowdsensing

Abstract

Train operation status is an important piece of information for our transportation plans. Without the latest train operation status, commuters might be face transportation inconvenience. Nowadays, train operation status is managed by railway companies. When a railway company delays the status update, commuters do not know the actual train operation status. In rural areas, the status updates are often delayed because railway companies focus more of their efforts on the recovery of train operation. Therefore, we propose a crowdsourced train detection system using a microphone on a smartphone. In our train detection system, a smartphone analyzes the frequency components of sound signals acquired by a microphone. We calculate the probability of a train passing using a logistic regression model on the sound frequency components and apply a hysteresis thresholding with two thresholds to detect passing trains. In addition, simple filtering based on train length is also applied to increase robustness to noise, including the sound of other passing vehicles. We conducted initial experimental evaluations and confirmed that our train detection system can successfully detected trains with an F-measure of 0.99 and a recall of 1.0. Further, we also conducted experiments in a more practical environment where the audio signals were acquired by smartphones in pants’s pockets, and confirmed that the acquired audio signals are useful for train detection.