Send the link below via email or IMCopy
Present to your audienceStart remote presentation
- Invited audience members will follow you as you navigate and present
- People invited to a presentation do not need a Prezi account
- This link expires 10 minutes after you close the presentation
- A maximum of 30 users can follow your presentation
- Learn more about this feature in our knowledge base article
Transcript of HCI
Humanity is witnessing a revolution in interacting with computers..
technology is completely integrated with every aspect of our modern life...common users have higher expectations every day.
how many times do we use our smart phones daily..and for reasons other than making a phone call
Researchers are currently investigating every possible input in order to completely eliminate the barriers between humans and computers..and achieve the ultimate interaction..
What do you think the next possible input will be
It's the human brain... this magnificent organ that fascinated researchers all over the world
Imagine that you can communicate with any computer or smart device with only your thoughts..how easy and intuitive would that be..
this revolutionary idea is called the Brain Computer Interface or the (BCI) and it is far beyond the proof of concept.
EEG Signal Acquisition:
Electroencephalogram are the electrical signals recorded from the scalp surface, after being generated within the brain due to chemical reactions between the neural cells.
Several research proved that the variation in EEG patterns reflects corresponding functional brain activities such as motor imagery or mental arithmetic tasks
Among various methods to measure brain activity, EEG have many advantages:
If several mental states can be reliably distinguished by recognizing patterns in EEG, then a person could
communicate to any device by composing sequences of these mental states.
 E. R. Kandel, J. H. Schwartz, and T. M. Jessell, Principles of Neural Science, 3rd ed.New York: Elsevier/North-Holland, 1991.
EEG is picked up by metal electrodes and conductive media non invasively
Advantages: Avoid surgical implantation of microelectrodes in invasive recording methods.
Disadvantages: low spatial resolution due to skull spatial smearing effect as opposed to invasive EEG recording.
Standard 10-20 electrode system
Each letter correspond to the underlying brain lobe, odd numbers are for the left hemisphere, and even numbers are for the right
EEG signal characteristics interesting for BCI
8 to 256 electrodes over the skull
long preparation time, suitable for laboratory experiments
There are 5 main brain waves distinguished by their frequency ranges
Brain Neural Computer Interface
(BNCI) For Communication and Control
Event Related Potentials
potential changes in the EEG that occur in response to a particular “event” or a stimulus - time locked
: EEG over the visual cortex varies at
the same frequency as the stimulating light
With feedback training, subjects learn to voluntarily control their SSVER amplitude, but requires long term training.
This positive deflection in the EEG occurs about 300 ms after the stimulus onset.
its amplitude is strongly related to the unpredictability of the stimulus
RP and the MRP:
These are a response to a desire to perform and imagine movement respectively and are prominent at highly localized areas of the brain related to these function (primary motor cortex).
the subject has to wait for
the stimulus as opposed to the spontaneous EEG
it's an amplitude attenuation/enhancement of certain EEG rythms, namely the cortical mu rhythm, which is the same frequency range as the alpha rhythm but recorded over the motor cortex.
ERD is due to blocking of mu activity just before and during the real or imagery movement while ERS is associated with ceasing to move
During physical and motor imagery of right and left limb movements, beta and mu bands ERD occurs predominantly over
the contralateral left and right motor areas .This facilitates reliable decoding of movement intentions associated with different spatiotemporal ERD/ERS into 2D control commands.
To extract the most discriminative characteristics from the EEG patterns of the electrophysiological phenomena/mental task of interest.
EEG signal is challenging:
High dimensional, noisy and nonstationary.
It is not enough to use simple methods such as a band pass filter to extract the desired band power.
Dimension reduction techniques such as principal component analysis (PCA) or independent component analysis(ICA) can be applied to reduce the dimension of the original data
Signals are divided into short segments and the parameters can be estimated from each segment. FFT performs very poorly with short data segments. Wavelet transform or adaptive autoregressive components are preferred to reveal the non-stationary time variations of brain signals.
Multiples features can be extracted from several channels and from several time segments before being concatenated into a single feature vector.
Feature selection optimization algorithms may be attempted with the aim of minimizing the number of features while maximizing the classification performance.
PCA can identify the orthogonal basis functions of a subspace, which contains most of the data variance.
The orthogonal axes (Principle Components) are the first m eigenvectors corresponding to the first m largest eigen values of the covariance matrix of the data where m << n.
Valuable noise and dimension reduction method. PCA requires that artifacts are uncorrlated with EEG signals
The covariance matrix of the data is computed as follows:
to transform any input data vector of n dimension to the new subspace of m dimension: multiply this vector with the matrix U of m chosen principle components.
Applying ICA algorithm to EEG signals derives independent source from highly correlated EEG signals statistically and without regard to the physical location or configuration of the source generators
X is assumed to be an instantaneous linear mixture of n unknown
components or sources S via the unknown mixing matrix A
The goal of ICA is to recover statistically independent sources given only sensor observations by estimating the unmixing matrix W that approximates to recover an approximated version S’ ,
of the original sources
each estimated component can be projected on the channel locations creating topographic maps which help studying the corresonding component
Parametric description: Autoregressive (AR) Model
In autoregressive (AR) modelling of signals each sample of a
single-channel EEG measurement is defined to be linearly related with respect to a number of its previous samples, i.e.
the model parameters are estimated by employing some iterative optimization techniques in order to minimize the error between the actual and predicted values
In an autoregressive moving average (ARMA) linear predictive model each sample is obtained based on a number of its previous input and output sample values, i.e.
In a multivariate AR (MVAR) multichannels scheme is considered. Therefore,each signal sample is defined versus both its previous samples and the previous samples of the other channels, i.e. for channel i,
Signal Transforms and Joint Time–Frequency Analysis
EEG is nonstatonary: FFT perform very poorly with short time segments due to spectral smearing
Parametric spectrum estimation methods such as those based on AR or ARMA modelling can outperform the FFT but they may suffer from poor estimation of the model parameters
A time–frequency (TF) approach is the solution to the problem.
The short-time Fourier transform (STFT) is defined as the discrete-time Fourier transform evaluated over a sliding window. The STFT can be performed as:
where the discrete-time index n refers to the position
of the window w(n).
perfect resolution cannot be achieved in both time and frequency domains.
Windows are typically chosen to eliminate discontinuities at block edges and to retain positivity in the power spectrum estimate.
The choice also impacts upon the spectral resolution of the resulting technique
The wavelet transform (WT) is another alternative for a time–frequency analysis.
Wavelet Packet decomposition :
The Wavelet Packet Transform (WPT) is one such time frequency analysis tools. It is a transform that brings the signal into a domain that contains both time and frequency information
Wavelet packet decomposition is a wavelet transform where the signal is passed
through more filters than the discrete wavelet transform (DWT).
The aim of the classification step in a BCI system is recognition of a user’s intentions on the basis of a feature vector that characterizes the corresponding brain activity
The design of the classification step involves the choice of one or several classification algorithms from many alternatives.
Several classification algorithms have been proposed such as k-nearest neighbor classifiers, linear classifiers, support vector machines, and neural networks, among others.
The general trend prefers simple algorithms to complex alternatives. Simple algorithms have an inherent advantage because their adaptation to the features of the brain signal is inherently simpler and more effective than for more complex algorithms.
Very simple classifier, commonly used in BCI experiments providing acceptable results
A good choice for designing online BCI systems with a rapid response
It can lead to completely erroneous classifications in the presence of outliers or strong noise
For a two-class problem, LDA assumes that the two classes are linearly separable. According to this assumption, LDA defines a linear discrimination function which
represents a hyperplane in the feature space in order to distinguish the classes.
In the case of an N-class problem (N > 2), several hyperplanes are used.
The decision plane can be represented mathematically as:
The input feature vector is assigned to one class or the other on the basis of the sign of.g(x)
ANNs are non-linear classifiers that have been used in many applications
One of the most well-known ANN structures is the multilayer perceptron (MLP) which have been used successfully to classify up to 5 mental tasks.
Besides MLP, different types of ANN architecture have been used in the design of BCI systems such as Probabilistic Neural Networks (PNN), Fuzzy ARTMAP Neural Networks, Finite Impulse Response Neural
Networks (FIRNN) or Probability estimating Guarded Neural Classifiers (PeGNC)
SVM is a classifier that, in a similar way to LDA classifiers, constructs a hyperplane or set of hyperplanes, in order to separate the feature vectors
into several classes. However, in contrast to LDA, SVM selects the hyperplanes that maximize the margins, that is, the distance between the
nearest training samples and the hyperplanes
The basis of SVM is to map data into a high dimensional space and find a separating hyperplane with the maximal margin according to Cover’s theorem on the separability of patterns
SVM with non-linear decision boundary can be created by means of a kernel function K(x, y). Non-linear SVM leads to a more flexible decision boundary in the data space which may increase classification accuracy
The kernel that is usually used in the BCI field is the Gaussian or Radial Basis Function (RBF):
SVM has been widely used in BCI, because it is a simple classifier that performs well and is robust with regard to the curse of dimensionality, furthermore, SVM is speedy enough for
Currently research interest in BCI is exploding all around the world, the main motive behind most of the research experiments is: Helping Severely disabled persons to live a better life and communicate with the outside world..
Stroke patients' nervous system can be very damaged to the extent they can not use their hands to grasp objects, operate a keyboard/mouse or even navigate a wheel chair..Being continuously depending on a caregiver may be very frustrating. Moreover, in the communication era we live in those people may be completely dissociated from society.
For more severely injured people suffering from ALS, patients can be completely locked in their own bodies while their minds are still intact, and or them the BCI is their only hope.
Currently amazing results were already achieved but mostly with invasive BCI systems: this video shows how a patient succeeded to feed herself "chocolate" with a robot arm only controlled with thoughts..
extensive research should be done in order to take non invasive BCI out of the laboratories and make it available to common users in society..
We believe that controlling computers with thoughts may be the new revolutionary wave in HCI, helping paralyzed people all around the world to live a better life and moreover, taking the HCI experience of healthy users u to the next level..
Currently, the Gaming and entertainment industry is very interested in the BCI research as it might add a new dimension of communication that can not be achieved using any other method..
Games can include options to engage players in a challenging exciting environment, controlling for example avatars in the virtual world.
The right hemisphere
controls the left side of the body,
In most people, the left brain seems dominant
in its control of language and logic, the right hemisphere in spatial perception, art, music
and creative thought
Wavelet Packet Decomposition over 3 levels. g[n] is the low-pass approximation coefficients, h[n] is the high-pass detail coefficients
Combining EEG with other physiological signals such as EOG to achieve better control
new EEG recording devices, with affordable cost and easy out of the laboratory setup should be investigated for best possible results
In fact, The application of BCI in Gaming,
is very important for people suffering from disabilities as well. In this case the games are termed: "
Games for health
", which is a new research area that gained momentum the last couple of years.
BCI and games for health play a very important role
that satisfy needs in health care sector and and in rehabilitation processes.
brain computer interface technology allows
patients to make their rehabilitation therapy through dynamic video games while the
specialist acquires data from EEG signals.
The General Architecture of the BNCI: