Inferring station numbers in metro trips using mobile magnetometer sensor via an unsupervised k-means clustering algorithm

Transport mode detection in urban areas with the help of mobile phones is not anymore, a challenging problem. Most studies via machine and deep learning models using GPS and inertial mobile sensors can distinguish different modes with various prediction accuracies. This paper presents a new way to count the number of stations in metro trips using a magnetometer sensor where GPS, internet, and wireless positioning are unavailable. The primary source for this investigation was recorded via mobile magnetometer sensor of metro riders. We first find contextual features that can effectively recognize acceleration state according to the 3D magnetometer data and then classification with a k-means unsupervised method into different classes. Finally, we present a station counter algorithm to count the number of metro stations in metro-based trips. Results from experiments in Rome and Stockholm metro systems show that our final algorithm can count the number of stations with an accuracy of 86 % where there is no internet, GPS, and WiFi access.