Near Real-Time Transportation Mode Detection Based on Accelerometer Readings

This paper describes a method for automatic transportation mode detection based on smartphone sensors. Our approach is designed to work in real-time as it only requires 5s of sensor readings for the detection. Because we used accelerometer instead of GPS signal it uses less battery power and is therefore more user and phone-friendly. For the mode detection we use multiple support vector machine models which enable us distinguishing between multiple modes (bus, train, car). Before the classification, raw measurements are preprocessed in order to cancel out the constant acceleration that is caused by the force of gravity. The results of the paper are promising and are based on the collected training data from approximately 20 hours of driving on trains and public buses in Ljubljana.