Information Technology Reference
In-Depth Information
In 2006, a new model was presented enabling
the development of sophisticated systems that
facilitate care giving and clinical assessment of
persons with dementia. The model uses the Web
Ontology Language to quantify agitation through
the Scale to Assess Observed Agitation in Persons
with Dementia of the Alzheimer Type (SOAPD)
(Fook, et al., 2006). The model requirement is two-
fold: to capture all the characteristics of context
information relating to agitation monitoring in
persons with dementia and intervene by process-
ing and relaying information in a context-aware
manner. The authors used the SOAPD developed
by Hurley et al. (1999) to classify the degree of
agitation experienced by a demented person. The
tool consists of seven scale items where each item
is an observable behavior. Every five minutes the
duration of the individual's bodily movements and
vocalizations is rated. For Repetitive Motions in
Place and Outward Motions, intensity of the be-
havior is also rated. Total score for an observation
session is derived as the sum of the weights of all
observed behaviors. With this system, which can
automatically determine agitation, arrangements
can be made for suitable therapeutic interven-
tions like relaxing music; if the individual fails
to respond to the intervention, another trigger is
alarmed to send a message to the caregiver of the
person with dementia that the person is becoming
agitated (Fook, et al., 2006). This work does not
describe or implement any actual device for the
detection of agitation.
•
The choice of the biological signals that
are monitored. Signal acquisition has to be
non-invasive and as transparent as possible
to the monitored subject. Hence a limited
number of bio-signals are available which
include skin temperature, galvanic skin
response, electrocardiogram (ECG), pupil
diameter and respiration.
•
The choice of the type of emotion that the
algorithm will predict. Emotions are con-
troversially defined, but a common set of
basic emotion labels are: sadness, happi-
ness, fear, anger, surprise and disgust, as
defined by Ekman (Ekman, 1984).
•
The choice of a safe method to induce these
types of emotions. The mostly used meth-
ods are video clips that contain a scene cor-
responding to the desired emotion, audio
clips composed of music or sounds that
correspond to the desired induced emotion,
or a mix of both methods.
•
The choice of the features that are ex-
tracted from the bio-signals as well as the
recognition algorithm that is used to detect
emotion. In general, the features extracted
are from the ECG which are commonly the
total energy, the RMS value and the inter-
beat interval. The recognition algorithms
include K-means or C-mean clustering,
regression algorithms such as canonical re-
gression, neural networks, support vector
regression and others.
Emotion Detection
ECG, skin temperature, skin conductance, and
respiration have been used to extract 22 features
for emotion recognition (Li, et al., 2006). The
induction of fear, joy and neutrality was done us-
ing video clip stimuli. The authors used canonical
correlation to achieve an 85.3% of correct clas-
sification ratio. This method can be improved by
reducing the number of features extracted from
the ECG and trying to omit respiration from the
monitored signals which can be uncomfortable
for the subject. Another work developed emotion
recognition using ECG, skin temperature and
Emotion detection has been shown to be an es-
sential tool in developing machine intelligence
(Picard, et al., 2001). In general an emotion detec-
tion experiment can be treated through measuring
different types of biological signals or by using
image processing techniques for facial expres-
sion analysis (Liao, et al., 2005). The scope of
this chapter is the use of biological signals for
emotion detection, which in general has four main
challenges.
Search WWH ::
Custom Search