A Usability Study of Physiological Measurement in School Using Wearable Sensors

Figures

• 

  Figure 1. Electrodermal activity (EDA) sensor used in the experiment (EdaMove 4).

• 

  Figure 2. Distribution of EDA and heart rate (HR) signals in the bin width of 0.5 μS and 0.5 bpm, respectively.

• 

  Figure 3. Results on the removal of transitional and inter-loss period. The effect of variations of the transitional and inter-loss period on the averaged normalized metrics is shown in (a). Examples of data portions that were marked as invalid periods are shown in (b,c).

• 

  Figure 4. Continuous decomposition analysis (CDA-Ledalab) results from EdaMove signals before and after filtering and from ground-truth ActiveTwo signal. The smaller box illustrates in finer details. Locations of the estimated phasic peaks are shown. Significant peaks with the estimated amplitude higher than 0.05 μS are highlighted by using larger marks. Note that the scale on the left-vertical axis is for EdaMove signals, while the scale on the right-vertical axis is for ActiveTwo signal.

• 

  Figure 5. Comparison of EdaMove signals before and after applying various filtering/smoothing techniques at two different time periods (a,b).

• 

  Figure 6. Effect of varying filter size from 2 to 4 s in steps of 0.5 s on the number and amplitude of phasic (remaining and discarded) peaks and on the attenuation of the amplitude of the peaks following emotional sounds (absolute values and relative values to the original peak heights in percent. Error bar represents SD/N where N = 12 participants.).

• 

  Figure 7. An example of EDA (in μS) of five participants in the same class. A red trace represents an original signal, a black trace represents the signal after applying S-G filter, and red shade represents the tonic level computed by CDA-Ledalab.

• 

  Figure 8. An example of artifactual (red) and questionable (blue-shade) segments as labeled by EDA-Explorer with the variation of the downscaling factors and rescaling to specific levels; (a) segments after downscaling (no scaling, 2, 4, 8) and (b) segments after rescaling to 16, 8, 4, 2 μS.

• 

  Figure 9. Features of EDA, HR, and accelerometer data over the course of 50-min class. A shade area represents SD/N where N = 85 participants for EDA and accelerometer and 80 participants for HR.

• 

  Figure 10. Normalized features of EDA, HR, and accelerometer data over the course of 50-min class. A shade area represents SD/N where N = 85 participants for EDA and accelerometer and 80 participants for HR.

• 

  Figure 11. Comparison of (absolute and normalized) features calculated from raw EDA signals (Raw), signals after removing invalid portions (zero-rmv), and signals after applying the Savitzky–Golay (S-G) filter (smoothed) for all participants. A shade area represents SD/N where N = 85 participants.

• 

  Figure 12. Comparison of (absolute and normalized) features calculated from raw EDA signals (Raw), signals after removing invalid portions (zero-rmv), and signals after applying the S-G filter (smoothed) for 11 participants whose signals were heavily contaminated by artifacts. A shade area represents SD/N where N = 11 participants.