Electrodermal Activity Sensor for Classification of Calm/Distress Condition

Abstract

This article introduces a new and unobtrusive wearable monitoring device based on electrodermal activity (EDA) to be used in health-related computing systems. This paper introduces the description of the wearable device capable of acquiring the EDA of a subject in order to detect his/her calm/distress condition from the acquired physiological signals. The lightweight wearable device is placed in the wrist of the subject to allow continuous physiological measurements. With the aim of validating the correct operation of the wearable EDA device, pictures from the International Affective Picture System are used in a control experiment involving fifty participants. The collected signals are processed, features are extracted and a statistical analysis is performed on the calm/distress condition classification. The results show that the wearable device solely based on EDA signal processing reports around 89% accuracy when distinguishing calm condition from distress condition.

Keywords: arousal; calmness; distress; electrodermal activity; valence; wearable.

Figure 1

Prototype of the wearable. (Left) The wearable is put in the wrist of the non-dominant hand. (Right) The electrodes are attached to the medial phalanges in the palm sides of index and middle fingers.

Figure 2

Electrodermal activity sensor building blocks. (a) Virtual ground. R${}_{d1}$ = R${}_{d2}$ = 100 k$\mathrm{\Omega }$, C${}_{d1}$ = 1 $\mathsf{\mu }$F, C${}_{b1}$ = C${}_{b2}$ = 100 nF, and C${}_b$ = 47 $\mathsf{\mu }$F; (b) Current source. R${}_{ref}$ = 825 k$\mathrm{\Omega }$; (c) Low-pass filter. R${}_p$ = 110 k$\mathrm{\Omega }$, C${}_p$ = 1 $\mathsf{\mu }$F, R${}_{a1}$ = 75 k$\mathrm{\Omega }$, $R_{a2}$ = 150 k$\mathrm{\Omega }$, C${}_{a1}$ = C${}_{a2}$ = 1 $\mathsf{\mu }$F, R${}_{g1}$ = R${}_{g2}$ = 10 k$\mathrm{\Omega }$, and C${}_f$ = 1 nF.

Figure 3

Different stages in EDA signal processing. (a) Raw EDA signal before filtering. (b) Raw EDA signal after low-pass filtering and stimuli onset. (c) Estimation of EDA baseline using a cubic spline approximation. (d) Skin conductivity response (SCR) obtained after the convolution process.