Internet of Things. IoT Infrastructures. Second International Summit, IoT 360° 2015, Rome, Italy, October 27-29, 2015, Revised Selected Papers, Part II

Research Article

Ultra-Low Power Context Recognition Fusing Sensor Data from an Energy-Neutral Smart Watch

Download
343 downloads
  • @INPROCEEDINGS{10.1007/978-3-319-47075-7_38,
        author={Michele Magno and Lukas Cavigelli and Renzo Andri and Luca Benini},
        title={Ultra-Low Power Context Recognition Fusing Sensor Data from an Energy-Neutral Smart Watch},
        proceedings={Internet of Things. IoT Infrastructures. Second International Summit, IoT 360° 2015, Rome, Italy, October 27-29, 2015, Revised Selected Papers, Part II},
        proceedings_a={IOT360},
        year={2017},
        month={6},
        keywords={Ultra-low power Smart watch Context recognition Machine learning Sensor fusion Energy neutral Feature selection},
        doi={10.1007/978-3-319-47075-7_38}
    }
    
  • Michele Magno
    Lukas Cavigelli
    Renzo Andri
    Luca Benini
    Year: 2017
    Ultra-Low Power Context Recognition Fusing Sensor Data from an Energy-Neutral Smart Watch
    IOT360
    Springer
    DOI: 10.1007/978-3-319-47075-7_38
Michele Magno,*, Lukas Cavigelli1,*, Renzo Andri1, Luca Benini,*
  • 1: ETH Zurich
*Contact email: magno@iis.ee.ethz.ch, cavigelli@iis.ee.ethz.ch, benini@iis.ee.ethz.ch

Abstract

Today sensors and wearable technologies are gaining popularity, with people increasingly surrounded by “smart” objects. Machine learning is used with great success in wearable devices and sensors in several real-world applications. In this paper we address the challenges of context recognition on low energy and self-sustainable wearable devices. We present an energy efficient multi-sensor context recognition system based on decision tree to classify 3 different indoor or outdoor contexts. An ultra-low power smart watch provided with a micro-power camera, microphone, accelerometer, and temperature sensors has been used to real field tests. Experimental results demonstrate both high mean accuracy of 81.5 % (up to 89 % peak) and low energy consumption (only 2.2 mJ for single classification) of the solution, and the possibility to achieve a self-sustainable system in combination with body worn energy harvesters.