05-04-2012, 12:10 PM
Emotion Assessment
10.1.1.72.6661.pdf (Size: 137.84 KB / Downloads: 58)
Introduction
Emotions pervade our daily life. They can help us
guide our choices, avoid a danger and they also play a
key role in non-verbal communication. Assessing
emotions is thus essential to the understanding of human
behavior. Emotion assessment is a rapidly growing
research field, especially in the human-computer
interface community where assessing the emotional state
of a user can greatly improve interaction quality by
bringing it closer to human to human communication.
Emotion expression and analysis
Emotions can be expressed via several channels and
various features can be analyzed to assess the
emotional state of a participant. Most studies focus on
the analysis of facial expressions or of speech ([5],
[6]). These types of signals can however (more or less)
easily be faked; in order to have more reliable emotion
assessments, we preferred to use spontaneous and less
controllable reactions as provided by physiological
signals. Physiological signals can be divided into two
categories: those originating from the peripheral
nervous system (e.g. heart rate, ElectroMyogram -
EMG, galvanic skin resistance-GSR), and those
coming from the central nervous system (e.g.
ElectroEncephalograms-EEG).
Data collection
This section details the creation of a database of
physiological features patterns and associated labels
corresponding to the underlying valence/arousal model
of emotions. This requires to elicit physiological
emotional responses, to define a precise protocol to
acquire the data and finally to extract relevant features.
Emotion elicitation
A prevalent method to induce emotional processes
consists of asking an actor to feel or express a
particular mood. This strategy has been widely used
for emotion assessment from facial expressions and to
some extent from physiological signals [8].
Acquisition protocol
We acquired data from 4 participants, 3 males, 1
female, aged from 28 to 49. One of the participants is
left handed. For EEG's we used a Biosemi Active Two
device [13] with 64 electrodes (plus 2 for reference).
The other sensors used were a GSR sensor, a
plethysmograph to measure blood pressure, a
respiration belt to evaluate abdominal and thoracic
movements, and a temperature sensor. All signals were
sampled at a 1024 Hz rate.
Classification
A Naïve Bayes classifier was first applied for each
participant. This classifier is known to be optimal in
the case of complete knowledge of the underlying
probability distributions of the problem. This is
unfortunately not the case in our study, since very few
samples are available to construct them;
Conclusion
In this paper two categories of physiological
signals, from the central and from the peripheral
nervous systems, have been evaluated on the problem
of assessing the arousal dimension of emotions. This
assessment was performed as a classification problem,
with ground-truth arousal values provided either by the
IAPS or by self-assessments of the emotion. Two
classifiers were used, Naïve Bayes or based on FDA.