Daily Routine Discovery from Wearable Sensors

Project start: 

01 June 2011

iCareNet Fellow: 

Abstract of the Project

The aging society introduces several difficulties into elderly care, one of them being the increasing number of elderly people to monitor on a regular basis. Extensive research is available regarding specific critical events such as fall detection. However, it might be feasible that tracking of daily routines allows reacting more robustly by detecting a changing trend in behavior which is connected to the person's health condition.

A first step towards an assisted living application is to develop appropriate models and algorithms to detect behavior patterns in daily live. As daily life activities are very subject dependent and vary regarding duration and individual activities involved, discovering structures in daily activities is a challenging research problem. For complex activity discovery often hierarchical models are used (Fig. 1). E.g. Huynh et al. [2] used probabilistic topic models to discover specific activity patterns from a number of primitives, which were mapped to complex daily routines.

This project focuses on the development of discovery algorithms such as topic models and their application in daily routine monitoring. Therefore we aim towards an improved activity recognition system  based on discovery which will be evaluated in simulated environment and real-world studies.








Fig.1: Hierarchical Model for Complex Activity Discovering

Research Achievements to date:

  • Defining scope and focus of the project in terms of methods and evaluation
  • Investigation on topic model performance stability in terms of dataset properties  [1]
  • Development of daily routine simulation model for dataset generation [1]
  • Definition of dataset requirements for stable model performance [1]


[1] Seiter, J.; Amft, O.; & Tröster, G.: Assessing topic models: how to obtain robustness?,Pervasive 2012 workshop proceedings, Pevasive 2012

[2] T. Huynh, M. Fritz, and B. Schiele. Discovery of activity patterns using topic models. In Proceedings of the 10th international conference on Ubiquitous computing, UbiComp ’08, pages 10–19, New York, NY, USA, 2008. ACM.