Abstract
This work is located in the studies of the brain and their signals. The puspose is to know when someone wants to make a movement. Thus, it might help to people that actually are not able to move a member of their body or more. Mainly, it is focused in the design of strategies
for detection of action potentials or spikes when a movement wants to be made. This study
is not looking for action potentials form, it is looking for patterns and characteristics that
allow to recognize the movement. Although there are action potentials covered by the signals
taken from the electrodes, but they are unavailable.
To accomplish the objective, it is used the Electroencephalogram (EEG) signals of a
public data base. It is selected the ones related to the movement of the hands, concretely,
the movement of open and close the fist. Signal sources of noise that dirty the signal are
analyzed, they are called artifacts, and then, filtering stage comes.
Now, possibles algorithms are checked. It is decided to use the Wavelet transform and the
way in which it obtains the energy of the signal. Thanks to the calculation of Wavelet energy
in 22 subjects, it is reached to the conclusion that Wavelet energy for movement is
higher than for not movement. So, electrodes that comply with this condition at 100%
are 4.
The final algorithm is implemented three features: correlation, a parameter that gives
a relation between two signals, the energy, their range, their standard deviation and their
average. It could be said that algorithm has two parts: a training part and a decision part.
Inside decision part, there are three algorithms: ProMove, ProMove + improve and Logic.
The basic difference among ProMove and Logic is an or (||) and an and (&). The improve
is based on empiric knowledge.
Final conclusions show that the signals between subjects are very changing.
Therefore, same algorithm is not useful for everybody. To some subjects, the successful
probability is very high (92,86% - 1 fail), while for others is more low than what is expected
(50% - 7 fail). With these test, the importance in the length of the signals is reflected,
because if signals for subjects with more than 3 fails are inversely processed, the fails are
reduced. The most useful algorithm for a larger number of subjects is ProMove + improve.