Advertisement

Activity Recognition

  • Subhasis Chaudhuri
  • Siddhartha Duttagupta
  • Tanmay D. Pawar
Chapter

Abstract

Wearable ECG recorders (W-ECG) provide a practical solution for ambulatory cardiac monitoring. W-ECGs are increasingly being used by people suffering from cardiac abnormalities, who also choose to lead an active lifestyle. From the discussions in the previous chapters regarding W-ECG, we can now understand that the challenge presently is that the ambulatory ECG signal is influenced by motion artifacts induced by body movement activity (BMA) of the wearer. The usual practice is to develop effective ltering algorithms which can eliminate the motion artifacts. However, due to spectral overlap between the motion artifact signal and the cardiac signal the complete elimination of the motion artifact from the ambulatory ECG signal is not possible without unduly aecting the cardiac signal component. Therefore, instead of filtering we would like to identify the presence of the motion artifact and the type of body movement from the analysis of the ambulatory ECG signal itself. We have already addressed the issue of detecting BMA transitions from the ECG signal in the previous chapter. The method proposed for the transition detection is an unsupervised learning approach which only looks for any abrupt changes in the nature of the motion artifact signal due to changes in BMA. However, a particular BMA is not yet characterized from the analysis of ECG in the previous chapter. In this chapter we focus on the BMA recognition from the ambulatory ECG signal for which we will use BMA classiers with certain specic types of BMA classes. The classication approach for BMA recognition requires supervised training of the specied BMA classes using the corresponding ECG data during the specied BMA. For this purpose we have recorded the ECG signals during specied BMA, e.g. sitting still, walking, movements of arms and climbing stairs, etc. with a single-lead W-ECG as described in Section 5.2. The collected ECG signal during the BMA is presumed to be an additive mix of signals due to cardiac activities, motion artifacts and sensor noise as per the mathematical model given in Section 1.5. We have successfully used the mathematical model of the ambulatory ECG in the previous chapter for the transition detection from the ambulatory ECG signal. Here we follow the same model of the ambulatory ECG signal for the analysis which leads to the recognition of dierent types of BMA from the ECG itself. According to the mathematical model, the motion artifact signal is one of the components of the ambulatory ECG signal which depends on the type of BMA and hence the BMA recognition should be possible from the analysis of the ECG signal. The motion artifact signal can be derived from the ambulatory ECG by suppressing the cardiac signal and the sensor noise. We hypothesize that a similar type of BMA induces a similar type of motion artifact whereas different types of BMA induce dierent types of artifact. If this is true then we can train a classier to detect the type of BMA class using the motion artifact signal. As per the mathematical model in Section 1.5, we first derive the motion artifact signal by estimating the cardiac signal. The derived motion artifact signal can be used for the BMA recognition. We use classiers trained for dierent BMA classes in which there are two types of representations: one is a nonparametric representation and the other is a parametric representation. In the nonparametric BMA classiers each of the BMA classes is represented by a set of vectors derived from the ambulatory ECG data for the specic BMA class during training. Whereas in the parametric BMA classiers, the individual BMA class is modeled by certain parameters derived from the ambulatory ECG data available for only that particular BMA class. Both kinds of representations obtained by the supervised learning are then used for classication of the ambulatory ECG signals to recognize the BMA class during testing. Here we use the derived motion artifact signal for supervised training of the BMA classiers and the classication of BMA types, which requires some preprocessing on the ambulatory ECG signals recorded by W-ECG. The details of preprocessing and analysis are presented in this chapter.

Keywords

Hide Markov Model Motion Artifact Activity Recognition Sensor Noise Gabor Feature 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

Copyright information

© Springer-Verlag US 2009

Authors and Affiliations

  • Subhasis Chaudhuri
    • 1
  • Siddhartha Duttagupta
    • 1
  • Tanmay D. Pawar
    • 1
  1. 1.Department of Electrical EngineeringIndian Institute of Technology, BombayMumbaiIndia

Personalised recommendations