Recognizing User’s Activity and Transport Mode Detection: Maintaining Low-Power Consumption

Abstract

Mobile phones are being used for more than just communication because of their wide range of capabilities in aspects of computation and sensing. In this paper, we propose an approach based on supervised learning to detect the user’s mode of transport based on the smartphone’s built-in accelerometer sensor and the location data. We create a convenient hierarchical classification system, proceeding from a coarse-grained to a fine-grained classification and no requirements of specific position and orientation setting is needed. This study explores how coarse-grained location data from smartphones can be used in combination with accelerometer data to recognize high-level properties of user mobility. Our approach can achieve over 95% accuracy for inferring various transportation modes including tram, bus, train, walking, and stationary. The results suggest that our approach of adding coarse-grained location data improves the accuracy of detection by 10% in comparison with the accelerometer only approach. We present a review of existing approaches for transport mode detection and compare them regarding the type of devices used as sensing unit, the sensors used, the considered transport modes, energy efficiency, and the algorithms used for the classification task.