Mobile Computing, Applications, and Services. Third International Conference, MobiCASE 2011, Los Angeles, CA, USA, October 24-27, 2011. Revised Selected Papers

Research Article

SensCare: Semi-automatic Activity Summarization System for Elderly Care

Download480 downloads
  • @INPROCEEDINGS{10.1007/978-3-642-32320-1_1,
        author={Pang Wu and Huan-Kai Peng and Jiang Zhu and Ying Zhang},
        title={SensCare: Semi-automatic Activity Summarization System for Elderly Care},
        proceedings={Mobile Computing, Applications, and Services. Third International Conference, MobiCASE 2011, Los Angeles, CA, USA, October 24-27, 2011. Revised Selected Papers},
        proceedings_a={MOBICASE},
        year={2012},
        month={10},
        keywords={Sensor-based Elderly Care Structured Activity Recognition Activity Summarization Lifelog},
        doi={10.1007/978-3-642-32320-1_1}
    }
    
  • Pang Wu
    Huan-Kai Peng
    Jiang Zhu
    Ying Zhang
    Year: 2012
    SensCare: Semi-automatic Activity Summarization System for Elderly Care
    MOBICASE
    Springer
    DOI: 10.1007/978-3-642-32320-1_1
Pang Wu1,*, Huan-Kai Peng1,*, Jiang Zhu1,*, Ying Zhang1,*
  • 1: Carnegie Mellon University
*Contact email: pang.wu@sv.cmu.edu, huankai.peng@sv.cmu.edu, jiang.zhu@sv.cmu.edu, joy.zhang@sv.cmu.edu

Abstract

The fast growing mobile sensor technology makes sensor-based lifelogging system attractive to the remote elderly care. However, existing lifelogging systems are weak at generating meaningful activity summaries from heterogeneous sensor data which significantly limits the usability of lifelogging systems in practice. In this paper, we introduce SensCare, a semi-automatic lifelog summarization system for elderly care. From various sensor information collected from mobile phones carried by elderlies, SensCare fuses the heterogeneous sensor information and automatically segments/recognizes user’s daily activities in a hierarchical way. With a few human annotations, SensCare generates summaries of data collected from activties performed by the elderly. SensCare addresses three challenges in sensor-based elderly care systems: the rarity of activity labels, the uncertainty of activity granularities, and the difficulty of multi-dimensional sensor fusion. We conduct a set of experiments with users carrying a smart phone for multiple days and evaluate the effectiveness of the automatic summary. With proper sensor configuration, the phone can continue to monitor user’s activities for more than 24 hours without charging. SensCare also demonstrates that unsupervised hierarchical activity segmentation and semi-automatic summarization can be achieved with reasonably good accuracy (average F1 score 0.65) and the system is very useful for users to recall what has happened in their daily lives.