IEEE Consumer Electronics Magazine - October 2016 - 46

ache, elevated blood pressure, and rapid heartbeat [5]. Chronic
stress can exhibit the same symptoms as acute stress, but with
more extensive damage. Chronic stress is identified as a risk
factor for hypertension and coronary disease [7], [8], irritable
bowel syndrome, gastroesophageal reflux disease [9], generalized anxiety disorder, and depression [10]. If chronic stress is
allowed to proliferate, ongoing symptoms may reduce the
quality of life for those experiencing this stress.
Early stress-detection methods relied heavily on self-reports
in response to a standardized list of questions. Examples of these
early methods include the Perceived Stress Scale 10 or 14 and the
Depression Anxiety and Stress Scale 21, which were based on
questionnaires regarding life events [15]. Although these methods have been validated, there is still the concern of subjective
response bias from the individual that may introduce additional
skewness into the assessment of stress. Hence, there is a need for
objective measures with a basis in physical and physiological
domains. Because stress also presents itself via biomarkers and
physical expression, it can be measured objectively and observed.
The classic standard for a physiological measure of stress
has been the assessment of cortisol levels produced by the
hypothalamic-pituitary-adrenocortical (HPA) axis. This assessment involved cortisol extractions from various sources: hair,
saliva, blood (serum), and urine. These measurements were
invasive and/or involved a laborious process for analysis [16],
[17]. As a result, the discovery of alternative means of measurement via wearable, as well as noncontact, sensors produced
resulted in alternatives that are now more commonly used.
In the following sections, predominant traditional and
contemporary methods for measuring stress via physiological
and physical means are discussed. Traditional methods
include ECG and EEG. However, the current trend in affective computing, particularly in stress detection, is leaning
more toward noninvasive methods of measurement dealing
with skin conductance and PPG (optical sensing).

PHYSIOLOGICAL STRESS-BASED
AFFECTIVE COMPUTING
Several indicators provide physiological measures of stress.
The most prevalent include heart activity (ECG), brain activity (EEG), skin response (GSR/EDA), blood activity (PPG),
respiratory response (piezoelectricity and electromagnetic
generation), and muscle activity (EMG). Each of these are
examined more closely in the following sections.

Heart activity and ecG

QRS complex, and T wave. The R spike of the QRS complex is
most commonly used for evaluation of HRV because it is the
most predominant spike in the waveform [21].
In stressful situations, HRV is a product of change in autonomic nerve activity, which is composed of sympathetic and
parasympathetic modulation. The function of the sympathetic
nervous system (SNS) as it relates to the heart is to speed HR to
provide an increase in blood supply to the body. This queues the
fight-or-flight rush that comes with stress. After the stressor has
been removed, the parasympathetic nervous system (PNS)
kicks in to slow HR. Examining the relationship of these two
nervous systems provides insight into the stress state of an individual [20], [18]; HRV provides such insight. The waveform of
a heart captures activity, and HRV further enables the analysis
of activity across multiple heart beats. This allows monitoring
of how quickly the body responds to stress, how long the stress
response lingers after the stimulus, and how rapidly the PNS
can act to reduce stress. When combined with other methods of
stress assessment, HRV can pinpoint corresponding physiological and physical markers associated with stress in an individual.

ecG
One of the most heavily used stress-detection methods uses ECG
(alternatively appearing as EKG) for heart activity such as HR
and HRV. ECG devices use electrodes that are strategically
placed on the body to measure electrical signals produced by
depolarization and repolarization of the heart [41]. There are a few
different configurations of this setup based on ECG lead polarity
and electrode placement. The most common configuration is
the standard limb, bipolar lead setup illustrated in Figure 1(a).
Figure 1(b) shows the Einthoven's triangle, which translates lead
placements to axial references in the electrical signals recorded
by each electrode.
HRV data extracted from ECG measurements can be filtered by frequency content and then analyzed. Frequency content is generally divided into two bands: low frequency (LF)
and high frequency (HF), with LF at 0.04-0.15 Hz and HF at
0.15-0.4 Hz [40]. It also may be common to see three-frequency bands comprised of LF (0-0.08 Hz), midfrequency (0.08-
0.15 Hz), and HF (0.15-0.5 Hz) [39]. Alternatively, these may
be labeled as very LF (VLF), LF, and HF, respectively. This
separation of frequency content is useful because SNS activity
is associated with LF content, whereas PNS activity is associated with HF content [40]. The analysis involved in assessing
HRV lies in the energy ratio of LF to HF content-in other
words, the ratio of SNS to PNS activity. This is usually modeled as [39]

Heart activity
A predominate factor examined for stress across many studies
involves heart rate (HR) and heart rate variability (HRV). When
stress is induced in an individual, HR becomes elevated. This is
part of the fight-or-flight response that is often referenced in
high-stress situations. HRV provides more information than HR
alone. HRV is the measure of standard deviation in interbeat
intervals of successive R waves in a heart beat [18]. A typical
heart beat consists of four main components: baseline, P wave,
46 IEEE Consumer Electronics Magazine

OCTOBER 2016

Power Ratio ECG = PowerLF .
PowerHF
The power values are typically extracted from power-spectral
density (PSD) which uses fast Fourier transforms (FFTs) to
convert time-domain data into frequency-domain data [40].
Alternatively, power densities may be normalized to total
power using the following models:

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - October 2016