4th International ICST Conference on Pervasive Computing Technologies for Healthcare

Research Article

User-friendly system for recognition of activities with an accelerometer

Download
  • @INPROCEEDINGS{10.4108/ICST.PERVASIVEHEALTH2010.8853,
        author={Gerd Krassnig and Daniel Tantinger and Christian Hofmann and Thomas Wittenberg and Matthias Struck},
        title={User-friendly system for recognition of activities with an accelerometer},
        proceedings={4th International ICST Conference on Pervasive Computing Technologies for Healthcare},
        proceedings_a={PERVASIVEHEALTH},
        year={2010},
        month={6},
        keywords={Acceleration, Accelerometers, Cardiac disease, Cardiovascular diseases, Classification tree analysis, Humans, Monitoring, Performance evaluation, Sensor systems and applications, Testing},
        doi={10.4108/ICST.PERVASIVEHEALTH2010.8853}
    }
    
  • Gerd Krassnig
    Daniel Tantinger
    Christian Hofmann
    Thomas Wittenberg
    Matthias Struck
    Year: 2010
    User-friendly system for recognition of activities with an accelerometer
    PERVASIVEHEALTH
    ICST
    DOI: 10.4108/ICST.PERVASIVEHEALTH2010.8853
Gerd Krassnig1,*, Daniel Tantinger1, Christian Hofmann1, Thomas Wittenberg1, Matthias Struck1
  • 1: Department of Image Processing and Medical Engineering, Fraunhofer Institute for Integrated Circuits lIS, Erlangen, Germany
*Contact email: krassngd@iis.fraunhofer.de

Abstract

Monitoring of a person's daily activities can provide valuable information for health care and prevention and can be an important supportive application in the field of ambient assisted living (AAL). The goals of this study are the classification of postures and activities using knowledge-based methods as well as the evaluation of the performance of these methods. The acceleration data are gained by a single tri-axial accelerometer, which is mounted on a specific position on the test subject. A data set for training and testing was gained by collecting data from subjects, who performed varying postures and activities. For these purposes, three different knowledge-based (decision tree and neural network) classification methods and a hybrid classifier were implemented, tested and evaluated. The results of the tests illustrated that the hybrid classifier performed best with an overall accuracy of 98.99%. The advantages of knowledge-based methods are the exchangeable knowledge base, which can be developed for different types of sensor positions and the state of health of the subject.