-
EVALUATION OF EMOTIONAL COMPONENTS TO IMPROVE SSVEP-BCI
Anibal Cotrina∗, Alessandro B. Benevides∗, Andre Ferreira∗,
Teodiano Bastos∗, MariaL. R. Menezes†, Carlos E. Pereira†, Javier
Castillo‡
∗Post-Graduate Program of Electrical Engineering, Federal
University of Espirito Santo (UFES),Av. Fernando Ferrari 514,
Vitoria, Brazil
†Post-Graduate Program of Electrical Engineering, Federal
University of Rio Grande do Sul (UFRGS),Av. Osvaldo Aranha 103,
Porto Alegre-RS, Brazil
‡Post-Graduate Program of Electronics Engineering, University
del Valle (UNIVALLE),Av. Paso Ancho 1300, Cali, Colombia
Emails: [email protected], [email protected],
[email protected],[email protected], [email protected],
[email protected],
[email protected]
Abstract— Brain-computer interface (BCI) provides a direct
connection between the user’s brain signalsand a computer,
generating an alternative channel of communication that does not
involve the traditional wayas muscles and nerves. Recent decades
have seen BCI applications as a novel and promising new channel
ofcommunication, control and entertainment for disabled and healthy
people. However, BCI technology can beprone to errors due to the
basic emotional state of the user: the performance of reactive and
active BCIsdecreases when user becomes stressed or bored, for
example. Passive-BCI is a recent approach that fuses BCItechnology
with cognitive monitoring, providing valuable information about the
user’s intentions, the situationalinterpretations and mainly the
emotional state. In order to improve the accuracy of BCIs, subjects
can performsimultaneous or sequential tasks typically used in two
BCI approaches in a hybrid condition that combinesboth BCIs. In
this work, a system composed of a passive-BCI co-working with a
reactive-BCI, with the aimof improving the performance of the
reactive-BCI is proposed. Thus the possibility of adjusting
recognitioncharacteristics of SSVEP-BCIs using a passive-BCI output
is evaluated.
Keywords— Emotional components, passive-BCI, reactive-BCI,
SSVEP-based BCI, asymmetry index
1 Introduction
A Brain-Computer Interface (BCI) provides a di-rect connection
between the user’s brain signalsand a computer, generating an
alternative chan-nel of communication that does not involve
thetraditional way as muscles and nerves (Wolpawet al., 2002). A
BCI defines a new input modal-ity for human-machine interaction
(HMI), whichcould substitute or add up to other input modal-ities
like manual input. Distinct mental statescan be associated with
physical actions, such assending the command “turn right” to a
wheelchairrobot just imagining the movement of the righthand
(Ferreira et al., 2010). Although presentingmany advantages, most
current BCIs are highlysusceptible to emotional states experienced
by itsusers, since emotions indicate what is importantand what you
care about (Picard, 2010). However,the BCIs have the advantage of
direct access tobrain activity, being able to provide meaningful
in-formation about the user’s emotional state. Suchinformation may
be used in two forms (Molinaet al., 2009): 1) Knowledge of the
influence of emo-tional state in the patterns of brain activity
allowsthe BCI to adapt their recognition algorithms sothat the
user’s intent is still interpreted correctlydespite signal changes
induced by the emotionalstate of the user. 2) The ability to
correctly rec-ognize emotions in BCIs that can be used to pro-vide
the user a more natural and intuitive way to
control the BCI in affective modulation. In thepresent work, a
passive-BCI able to co-work witha specific reactive-BCI, is
evaluated in order toimprove the performance of the BCI by
evaluat-ing emotional components. Experimental resultsare shown and
the proposal seems effective.
1.1 BCI categorization
According to the categorization proposed in(Zander and Kothe,
2011), active-BCIs have out-puts derived from brain activity, which
is directlyand consciously controlled by the user, thereforebeing
independent of external events (Wolpawet al., 1991); and
reactive-BCIs have outputs de-rived from brain activity arising in
reaction toexternal stimulation, which is indirectly modu-lated by
the user (Muller, Celeste, Bastos andSarcinelli, 2010).
Passive-BCIs have outputs de-rived from implicit information on the
actual usermental state, which arises arbitrarily without
thepurpose of voluntary control. The first two cate-gories derive
their outputs for controlling an ap-plication and the last one
derive its output to im-prove human-environment interaction or
human-machine interaction.
1.2 Reactive-BCI based on SSVEP
An event related potential (ERP) used in manyBCI systems is the
visual evoked response (VEP).
Anais do XX Congresso Brasileiro de Automática Belo Horizonte,
MG, 20 a 24 de Setembro de 2014
134
-
PASSIVE BCI(Emotion)
REACTIVE BCI(SSVEP) BCI command
FlickeringStimulus
BCI user Hybrid-BCI
Figure 1: Schematic overview of a passive-reactive Hybrid
BCI.
This potential, occurring involuntarily in responseto a visual
stimulus, can be measured over oc-cipital brain areas. Steady-State
VEP (SSVEP)is a periodic response elicited by repetitive
pre-sentation of a visual stimulus, with the samefundamental
frequency as that of the flickeringstimulus as well as its
harmonics (Middendorfet al., 2000), (Sutter, 1992), (Muller, Bastos
andSarcinelli, 2010). In a typical SSVEP-based BCIsystem, multiple
stimuli flickering at different fre-quencies are shown to the
subject. The increasein the SSVEP amplitude can be detected in
theelectroencephalographic (EEG) signal, which arefurther
processed, classified and translated intocontrol commands (Wang et
al., 2006), (Gaoet al., 2003), (Cheng et al., 2002), (Muller-Putzet
al., 2005).
1.3 Passive-BCI based on Emotion Components
A Passive-BCI is a recent approach that fuses BCItechnology with
cognitive monitoring, providingthe computer information about the
user’s inten-tions, the situational interpretations and mainlythe
emotional state. Emotions can be defined asa subjective, conscious
experience characterizedprimarily by psycho-physiological
expressions, bi-ological reactions, and mental state (Kleinginnaand
Kleinginna, 1981). Affective computing stud-ies techniques that
recognize, interpret, and pro-cess human emotions (Picard, 2003).
Asymmetryof the frontal lobe, given by the variation of the al-pha
band power of the EEG signals, is significantlyassociated with
human emotional states; in which,high alpha band power in the right
hemisphere isassociated to negative emotional states while
highpower in the left hemisphere is associated withpositive
emotional states (Davidson, 1992).
1.4 Hybrid BCI
In order to improve the accuracy of BCIs, subjectscan perform
simultaneous or sequential tasks typ-ically used in two BCI
approaches. A hybrid BCIis assembled by a collection of systems
that worktogether to provide a communication pathway be-tween the
human brain and a computer (machine).A hybrid BCI based on two
different could com-bines active, reactive, and passive BCIs.
1.5 Assessment
Recently, a new perspective on BCI has emerged(Nijboer et al.,
2009), which suggests that notonly voluntary self-regulated signals
can be usedas input, but also involuntary signals might tellus
something about the state of the BCI user (e.g.the emotional and
cognitive state). It is assumedthat relevant features from these
involuntary sig-nals (also referred to as passive signals) can
beextracted and used to adapt the recognition al-gorithms of the
BCI. In sum, the knowledge ofthe emotional state influence in brain
activity pat-terns allows the BCI to adapt its recognition
algo-rithms with the aim that the user intentions wouldbe
interpreted efficiently.
In the present work, a passive-BCI monitorsemotional component
of the BCI user with theaim improving a SSVEP-BCI performance is
eval-uated. The increase of the SSVEP amplitude canbe detected in
the EEG signals and translated intocontrol commands. However,
stimuli flickeringcould cause a stress-related emotional state or
lossof attention, as reported in (Muller, Bastos andSarcinelli,
2010). In order to accommodate thisissue, we propose a system in
which passive-BCIco-works with a SSVEP-BCI, whose schematicoverview
is shown in Figure 1. The SSVEP-BCIdetects the elicited evoked
potential from EEGsignals registered at occipital electrodes. At
thesame time, the passive-BCI identifies emotionalcomponents of
user mental state from EEG signalson the frontal brain region. The
system is thenswitched to a ”stress mode” when specific com-ponent
of emotional state, like stress, is detectedand consequently the
success rate of SSVEP de-creases. In this mode, the passive-BCI
outputmodules the reactive-BCI characteristics aimingto maintain
the success rate.
EEG signals of one subject were employed.Two flickering stimuli
were used to evoke theSSVEP potential. Spectral density of signal
iscomputed using Hilbert Transform. Two ways ofbecome the SSVEP
more robust were evaluated:adjusting the amplitude response and
adjustingthe frequency response. Asymmetry index com-puted of the
alpha band from frontal electrodeswere used to evaluate the
emotional state of thesubject.
Anais do XX Congresso Brasileiro de Automática Belo Horizonte,
MG, 20 a 24 de Setembro de 2014
135
-
2 Methods
2.1 Subjects
Due to the preliminary aspects of this work, theevaluation was
performed with only one volunteerwith no previous history of
neurological or psy-chiatric disorder. The experiment was taken
withthe understanding and written consent of the sub-ject, who gave
informed consent. This study wasapproved by the research ethics
committee of theFederal University of Espirito Santo (Brazil).
2.2 Stimulus
Two stimuli, emitted by two 5 × 7 LED arrange-ments flickering
at 5.6 Hz and 6.4 Hz were dis-played simultaneously. The subject
seated in frontof the SSVEP box and was asked to gaze on thetarget
LED for 17 s after a beep tone, then askedto close his eyes for 5
s, ending the trial after a sec-ond beep tone. The EEG signal was
recorded be-tween seconds 5 and 17 of the trial. Two sessionsof 10
trials were performed during the experiment.
2.3 Signal acquisition
BrainNet36 (BNT) was the device used for EEGacquisition with a
cap of integrated wet electrodes.EEG signals from 19 electrodes
positioned accord-ing to the international 10-20 system were
regis-tered (Figure 2). The grounding electrode waspositioned on
the subject forehead and the bi-auricular reference was adopted.
The EEG wasacquired at a sampling rate of 200 Hz. BNT is adevice
for clinical purposes that does not exportdata in on-line mode.
Therefore, a TCP-IP basedsniffer programmed in ANSI C was developed
toexport these data, allowing the on-line processing,which was
performed on MATLAB.
Figure 2: International 10-20 system for electrodeplacement.
2.4 Preprocessing
Signals were filtered employing an elliptic band-pass (4 Hz - 50
Hz). Signals from O1 and O2 elec-trodes were used to verify the
SSVEP responses;other channels were employed to perform common
average reference (CAR) spatial filtering, in orderto reduce the
correlation between channels origi-nated by external noise. CAR
filter is given by:
µCARi = µERi −
1
n
n∑j=1
µERj , (1)
where µCARi is the filtered signal and µERi is the
potential between the i-th electrode and the ref-erence
electrode.
2.5 Spectral density of an analytical signal
In rhythm modulation-based BCIs, the input of aBCI system is the
modulated brain rhythms withembedded control intentions. Brain
rhythm mod-ulation is realized by executing task-related
activi-ties, e.g., attending to one of several visual
stimuli.Demodulation of brain rhythms can extract theembedded
information, which will be convertedinto a control signal. The
brain rhythm modu-lations could be sorted into the following
threeclasses: power modulation, frequency modulation,and phase
modulation. For a signal s(t), its an-alytical signal g(t) is a
complex function definedas:
g(t) = s(t) + jŝ(t), (2)
where ŝ(t) is the Hilbert transform of s(t), definedas:
ŝ(t) =1
π
∫ ∞−∞
s(t)
t− τdτ . (3)
Due to the ŝ(t) have the same energy as s(t), en-ergy spectral
density is given by:
P (f) =1
4Ĝ(f)Ĝ(f)∗, (4)
where G(f) is the Fourier transform of g(t) andĜ(f)∗ denotes
the complex conjugate of G(f).The analytic signal has no power at
negative fre-quencies.
2.6 Adjusting of Amplitude of the Response
The amplitude of the SSVEP response of the EEGsignals depends on
the quantity of samples em-ployed to perform the FFT transform.
Normal-ized amplitude spectrum is calculated by:
Pnorm(f) =P (f)∑P (f)
, (5)
where P (f) is spectral energy density of the ana-lytical
signal.
∑(Pf) denotes a summation over
the total frequency points of a spectrum. The re-sponse becomes
more robust when more samplesare considered.
Anais do XX Congresso Brasileiro de Automática Belo Horizonte,
MG, 20 a 24 de Setembro de 2014
136
-
2.7 Adjusting the Frequency of the Response
The frequency corresponding to amplitude ofpeaks in the
frequency domain is compared withstimuli flickering frequencies to
determine whichstimulus was chosen by the subject. However, itis
common that the frequency of the peak (fun-damental or harmonic) is
slightly different to thestimulus frequency, or other peaks appear
at fre-quency domain. To solve this problem, PowerSpectral Density
Analysis (PSDA), which involvesprocessing in the frequency domain,
was usedto perform automatic recognition of SSVEP re-sponses of the
target stimulus.
If there is a peak in the same frequency ofthe stimulus, the
error will be zero. If the error isnot zero, the ratio will be
small if the amplitude ishigh, and at different frequencies, the
error will besmall. In this case, fk and fh could be adjusted.The
power spectral density analysis around thestimulus frequency is
given by:
Sk =mP (fk)
m/2∑i=−m/2
P (fk + ifr)
, (6)
usually expressed in dB; m is number of samplesaround the
stimulus frequency, and fr is the fre-quency resolution which
depends on the Fouriertransformation. P (fk + ifr) is the power
den-sity around the stimulus frequency. In this studym = 60 was
considered.
So, given k-th stimulus frequency fk, thecloser peak response
frequency fh, and the mag-nitude of the peak frequency P (fk), the
followingratio of proportion was used:
Ratio =|fk − fh|P (fk)
. (7)
2.8 Asymmetry index
The index of asymmetry of alpha band can becomputed by comparing
the power of contra-lateral frontal electrodes, in order to
identify com-ponent of stress-related emotional states.
Frontalcortex asymmetry has provided evidence thatgreater right
frontal activity seems to be morehighly related to negative
emotional states. Thisindex, that has a value between -1 and 1,
canbe employed as a switch to shift the system tothe ”stress mode”
(Davidson, 1992). The mostcommonly reported of the indexes is
computed bysubtracting the left hemisphere alpha power (Plh)from
the right hemisphere alpha power (Prh):
Assymetry =Plh − PrhPlh + Prh
,
where Plh and Prh were estimated by computingthe Power Spectral
Density.
3 Results
As mentioned above, the subject was asked tochoose one specific
target between two stimuliflickering at 5.6 Hz and 6.4 Hz. A
particularmental state, such as stress, can affect the fre-quency
or the amplitude of this potential. There-fore, a technique based
on adjusting the numberof samples employed to perform the FFT
trans-form and/or a technique based on the enlarge theratio of
searching of the peak response to com-pensate the frequency and
amplitude of evokedpotentials, respectively. Hence, elicited
SSVEPpotential response and asymmetry index when thesubject was
stimulated emotionally are presentedint his section.
3.1 Elicited SSVEP potential results
Hilbert transform was used to compute theSSVEP spectral
responses shown in the Figures3 and 4.
(a)
(b)
Figure 3: Normalized amplitude spectra corre-sponding to
different length. (a) For the stimu-lus flickering with 5.6 Hz. (b)
For the stimulusflickering with 6.4 Hz.
Figures 3(a) and 3(b) show the normalizedamplitude spectra
corresponding to four differentdata lengths. If the data length is
n = 200 sam-ples, corresponding to 1 s of signal; then, the
am-plitude of the response is weak. The response be-comes more
robust when more samples are consid-ered. Thus, SSVEP response peak
will be strongwhen n = 800 samples that corresponds to 4 s
ofsignal. Hence, responses that were computed withfew data are
affected with changes in the subjectmental states. In this sense,
one way to maintain
Anais do XX Congresso Brasileiro de Automática Belo Horizonte,
MG, 20 a 24 de Setembro de 2014
137
-
(a)
(b)
Figure 4: Normalized amplitude spectra SSVEPresponses of ten
trials (gray) and their the average(black).(a) For the stimulus
flickering with 5.6 Hz.(b) For the stimulus flickering with 6.4
Hz.
the SSVEP potential amplitude could be achievedthrough adjusting
the data length of each trial.The number of samples increases the
data and theprocessing time, so this assessment maintains
thesuccess rate, but reduces the information transferrate (ITR). In
practice, the length of the samplesare determined by a window
function.
Figures 4(a) and 4(b) show the normalizedamplitude spectra of
the SSVEP response of theelectrode O2 corresponding to ten trials
(graycurves) and the average curve (black curve).The response to
stimulus flickering with 5.6 Hzpresents weaker amplitude than its
second har-monic frequency (11.2 Hz); concurrently the peakat 5.6
Hz is wider. In cases like this, thepeak detection using a
threshold becomes ineffi-cient. Thus, adjusting of frequency of the
responsemethod can be a good alternative. On the otherhand, for 6.4
Hz the SSVEP response presentspeaks at fundamental and second and
third har-monic frequencies.
Topographic maps of all position of the sub-ject’s scalp is
showed in the Figure: The two stim-ulation frequencies (5.6 Hz and
6.4 Hz) and thesecond (11.2 Hz and 12.8) and third (16.8 Hz and19.2
Hz) harmonics of each. Note that these mapspresent data derived
with different frequencies. Itis clear from both stimulation
frequencies the to-pographies show occipital activity
characteristic ofSSVEP. Only the topography of second
harmonicpresents occipital activity on the right side. Fi-nally,
topography of the third harmonic presentsslight occipital
activity.
(a) (b)
Figure 5: Topographic maps of SSVEP responseover six
frequencies. (a) 5.6 Hz with their secondand third harmonics. (b)
6.4 Hz with their secondand third harmonics.
3.2 Asymmetry Index results
The alpha power of the contralateral electrodes F3and F4 was
estimated by computing the PowerSpectral Density based on the
modified peri-odogram. The absolute value of Fast
FourierTransformation provides the amount of informa-tion contained
at a given frequency, and the squareof the absolute value is
considered the power of thesignal. In order to compute the
asymmetry of al-pha band power at frontal lobe from
contralateralelectrodes F3 and F4, One-minute signal was
con-sidered to the analysis. One-second segments thetrial signal
were taken into account to perform theanalysis; thus, for each
trial, the CAR filtering,the wavelet frequency band decomposition,
andthe frequency band power estimation were per-formed. Computing
the asymmetry was realizedwhen subject was listening to unpleasant
soundswhile he was asked to gaze at an SSVEP stimu-lus. Four types
of sound stimuli were selected toelicit an emotional state: 1)
Nothing, 2) ruler ona bottle, 3) dental drill, and 4) baby
laughing. In(Kumar et al., 2012), stimuli (2) and 4) were ratedas
one of the most unpleasant sounds and the ofthe least unpleasant
sounds, respectively.
The box plot was used to show the distribu-tion of results (See
Figure 6). It can be seen that,in all cases the value of the median
of the indexwas negative. However, a clear difference betweenthe
sounds 2) and 4) is showed. It is evident thatthe results shows
that the sound 3) has the leastindex for the subject. It indicates
that a stress re-
Anais do XX Congresso Brasileiro de Automática Belo Horizonte,
MG, 20 a 24 de Setembro de 2014
138
-
lated emotional state was elicited on the subject,because high
alpha band power in the right hemi-sphere is associated to negative
emotional stateswhile high power in the left hemisphere is
asso-ciated to positive emotional states. Finally, theeffect of the
stimulus 4) was the same that the“stimulus nothing”.
Figure 6: boxplot with 90th percentile (10% and90%) of the
results of asymmetry index comput-ing.
4 Conclusion
The method of recognizing the fundamental fre-quency of an SSVEP
elicited response described inSections 2.6 and 2.7 can maintain the
error rate byadjusting two parameters fk and fh, that deter-mine
the window width around the stimulus fre-quency. Thus, it can be
concluded that the search-ing limits of evoked potential peaks and
the num-ber of samples used to compute the FFT transfor-mation can
be adjusted to improve the search ofthe SSVEP potential’s
frequency. Those resultsare promising because they show that
passive-BCIs could improve or maintain the accuracy ratedespite of
BCI user’s emotional states, such asstress. In the Section 2.8,
although the assessmentreduces the information transfer rate, it
maintainsthe error rate of the reactive-BCI. Since the asym-metry
or energy in alpha band can be used toidentify emotional components
of the BCI user,the next step in this work will be to integrate
thepassive-BCI and the reactive-BCI showed in theFigure 1 in order
to develop a more robust BCI.
The index is used to modulate the reactive-BCI characteristics
by using two adjustable pa-rameters using a specific emotional
componentsuch as asymmetry index: 1) the number of sam-ples to
compute the FFT Transform and 2) thesearch range around the
stimulus frequency at thespectral domain. Although the ITR
decreases be-cause the first adjustment increases the samplesand
the time between two trials, and the secondadjustment increases the
SSVEP peak searching
time, this assessment could improve the interac-tion between the
user and the reactive-BCI be-cause it maintains the success
rate.
Alpha power has been found to be more re-liably related to task
performance compared toother frequency bands, when the tasks
comparedcarefully match on psychometric properties. Al-pha power
asymmetry may be considered a gra-dient of power that exists
between the two ho-mologous electrodes in the pair, with the slope
ofthe gradient being towards the electrode with thegreatest amount
of power in this frequency band.
The next step in this research will be to com-pute the asymmetry
index and to propose a linearequation that ties in this index with
SSVEP-basedBCI parameters. It is well known that BCIs,
likeSSVEP-based BCIs, are not suitable for all users(Guger et al.,
2012). The causes for this ineffi-ciency have not yet been
satisfactorily described.Few studies exist that explicitly
investigated thepredictive value of internal (user related) and
ex-ternal (BCI related) factors on the BCI perfor-mance. The
accuracy of SSVEP can be monitoredby a Reclassifier, which evaluate
a number of con-secutive results. The Re-classifier is able to
acti-vate a switch if the accuracy is not being recov-ered. In this
case, an autonomous interface can beimplemented in order to take
control of the ma-chine. Commands like “Stop the machine”, “Re-turn
to previous stage” or “Return to the startingpoint” can be sent to
a control system, as shownin the Figure 7.
Re-Classifier
AUTONOMOUSINTERFACE
Switch
HYBRID-BCI
Return
Wait
Stop
Call
Figure 7: Schematic overview human-machine in-terface composed
by a passive-reactive hybrid BCIand an autonomous interface.
5 Acknowledgment
The authors would like to acknowledge the fi-nancial funding
from FAPES/CNPq (Process53666038/2011), and thank CAPES agency
forscholarship support.
Anais do XX Congresso Brasileiro de Automática Belo Horizonte,
MG, 20 a 24 de Setembro de 2014
139
-
References
Cheng, M., Gao, X., Gao, S. and Xu, D.(2002). Design and
implementation of abrain-computer interface with high
transferrates, IEEE Transactions on Biomedical En-gineering 49.
Davidson, R. (1992). Anterior cerebral asymme-try and the nature
of emotion., Brain andcognition 20(1): 125–151.
Ferreira, A., Bastos-Filho, T., Sarcinelli-Filho,M., J.L.M., S.,
Garcia, J. and Quintas, M.(2010). Improvements of a
Brain-ComputerInterface Applied to a Robotic Wheelchair,p. 64.
Gao, X., Xu, D., Cheng, M. and Gao, S. (2003).A BCI-based
environmental controller for themotion-disabled., IEEE transactions
on neu-ral systems and rehabilitation engineering :a publication of
the IEEE Engineering inMedicine and Biology Society 11(2):
137–140.
Guger, C., Allison, B., Growindhager, B., Pruckl,R.,
Hintermuller, C., Kapeller, C., Bruck-ner, M., Krausz, G. and
Edlinger, G. (2012).How many people could use an SSVEP
BCI?,Frontiers in neuroscience 6.
Kleinginna, P. R. and Kleinginna, A. M. (1981).A categorized
list of emotion definitions, withsuggestions for a consensual
definition, Moti-vation and Emotion 5.
Kumar, S., von Kriegstein, K., Friston, K.and Griffiths, T. D.
(2012). Features ver-sus feelings: Dissociable representations
ofthe acoustic features and valence of aver-sive sounds, The
Journal of Neuroscience32(41): 14184–14192.
Middendorf, M., G, M., Calhoun, G. and Jones,K. (2000).
Brain-computer interfaces basedon the steady-state visual-evoked
response.,IEEE transactions on rehabilitation engi-neering : a
publication of the IEEE En-gineering in Medicine and Biology
Society8(2): 211–214.
Molina, G. G., Nijholt, A. and Twente, U. (2009).Emotional
Brain-Computer Interfaces, 3rdInternational Conference on Affective
Com-puting and Intelligent Interaction and Work-shops, 2009. ACII
2009. .
Muller-Putz, G., Scherer, R., Brauneis, C. andPfurtscheller, G.
(2005). Steady-state vi-sual evoked potential (SSVEP)-based
com-munication: impact of harmonic frequencycomponents., Journal of
neural engineering2(4): 123–130.
Muller, S. M. T., Bastos, T. F. and Sarcinelli, M.(2010).
Incremental ssvep analysis for bciimplementation, 32nd Annual
InternationalConference of the IEEE EMBS, pp. 3333–3336.
Muller, S. M. T., Celeste, W. C., Bastos, T. F.and Sarcinelli,
M. (2010). Brain-computerinterface based on visual evoked
potentialsto command autonomous robotic wheelchair,Journal of
Medical and Biological Engineer-ing 30(6): 407–415.
Nijboer, F., Carmien, S., Leon, E., Morin, F.,Koene, R. and
Hoffmann, U. (2009). Affec-tive brain-computer interfaces:
Psychophysi-ological markers of emotion in healthy per-sons and in
persons with amyotrophic lat-eral sclerosis, Affective Computing
and In-telligent Interaction and Workshops, 2009.ACII 2009. 3rd
International Conference on,pp. 1–11.
Picard, R. (2010). Affective computing: Fromlaughter to IEEE,
Affective Computing,IEEE Transactions on 1(1): 11–17.
Picard, R. W. (2003). Affective computing:challenges,
International Journal of Human-Computer Studies 59.
Sutter, E. E. (1992). The brain response in-terface:
communication through visually-induced electrical brain responses,
Journal ofMicrocomputer Applications 15.
Wang, Y., Wang, R., Gao, X., Hong, B. andGao, S. (2006). A
practical VEP-based brain-computer interface., IEEE transactions
onneural systems and rehabilitation engineering: a publication of
the IEEE Engineering inMedicine and Biology Society 14(2):
234–239.
Wolpaw, J., Birbaumer, N., Dennis, M.,Pfurtscheller, G. and
Vaughan, T. (2002).Brain-computer interfaces for communica-tion and
control., Clinical neurophysiology :official journal of the
International Federa-tion of Clinical Neurophysiology 113(6):
767–791.
Wolpaw, J., D, M., Neat, G. and Forneris, C.(1991). An EEG-based
brain-computer inter-face for cursor control.,
Electroencephalogra-phy and clinical neurophysiology 78(3):
252–259.
Zander, T. and Kothe, C. (2011). Towards passivebrain-computer
interfaces: applying brain-computer interface technology to
human-machine systems in general., Journal of neu-ral engineering
8(2).
Anais do XX Congresso Brasileiro de Automática Belo Horizonte,
MG, 20 a 24 de Setembro de 2014
140