By train or by car? Detecting the user's motion type through smartphone sensors data

Nowadays, the increasing popularity of smartphone devices equipped with multiple sensors (e.g. accelerometer, gyroscope, etc) have opened several possibilities to the deployment of novel and exciting context-aware mobile applications. In this paper, we exploit one of this possibility, by investigating how to detect the user motion type through sensors data collected by a smartphone. Our study combines experimental and analytical contributions, and it is structured in three parts. First, we collected experimental data that demonstrate the existence of specific sensors data patterns associated to each motion type, and we propose methods for data analysis and features extraction. Second, we compare the performance of different supervised algorithms for motion type classification, and we demonstrate that jointly utilizing the multiple sensor inputs of a smartphone (i.e. the accelerometer and the gyroscope) can significantly improve the accuracy of the classifiers. At the same time, we analyze the impact of sampling parameters (e.g. the sampling rate) on the system performance, and the corresponding trade-off between classification accuracy and energy consumption of the device. Third, we integrate the motion type recognition algorithm into an Android application, that allows to associate a specific smartphone configuration to each detected motion type, and to provide this information at system-level to other context-aware Android applications. Experimental results demonstrate the ability of our application in detecting the user's motion type with high accuracy, and in mitigating the classification errors caused by random data fluctuations.