Battery-Efficient Transportation Mode Detection on Mobile Devices

The ubiquitous presence of sensor-rich smartphones offers excellent opportunities to capture and monitor individuals' contexts and activities such as the utilized modes of transport. Most of the current systems inferring transportation modes involve continuous sensing of GPS or acceleration modules, which inconveniently decreases the battery lifetime. This paper presents a more battery-efficient approach for transportation mode detection. This approach introduces an additional component able to detect whether a user is using any transportation mean or none at all, based on cellular network information only. Using this a priori knowledge, our inference model is able to avoid the battery-exhausting transportation mode detection when users are not travelling. Based on experiments involving six participants we show that our approach allows the transportation modes car, train and walk to be distinguished with an accuracy of 95.7%. Tests on a dataset collected by nine participants show that the a priori knowledge can be obtained with an accuracy of 85%. Further experiments reveal that our approach is able to gain 75% of battery lifetime. Finally, we outline an implementation of our detection approach providing flexible means to configure the use of smartphone sensors and thus supporting the kind of experiments presented in this paper.