Authors: Huiyu Sun; Suzanne McIntosh; Bin Li
Addresses: Department of Computer Science, New York University, New York, NY 10012, USA ' Department of Computer Science, New York University, New York, NY 10012, USA ' Department of Computer Science, New York University, New York, NY 10012, USA
Abstract: Smartphones are equipped with embedded sensors, which have been widely used for human activity recognition, context monitoring, and localisation. In this paper, we use smartphone sensors to detect in-progress phone calls and classify the caller's activity states. Phone call detection has a number of practical uses: it can be used to monitor activities at places where calls are forbidden and it can be used to assist activity recognition schemes. We propose a real-time phone call detection scheme using smartphone proximity and orientation sensor data and design Android applications to record, upload and display sensor data. We classify the caller's activity states into three categories: sitting/standing, lying down, and walking. Features are extracted from proximity and orientation sensors to train classifiers using different classification algorithms (Naive Bayes, logistic regression, and support vector machine). Experiments show our system achieves an overall accuracy of 91% in successfully detecting and classifying phone calls.
Keywords: human activity recognition; smartphone sensors; phone call detection.
International Journal of Sensor Networks, 2017 Vol.25 No.2, pp.104 - 114
Received: 28 Aug 2016
Accepted: 13 Oct 2016
Published online: 03 Oct 2017 *