Desarrollo de un sistema BCI basado en redes neuronales y movimientos de la cabeza para el manejo de un ordenador

Abstract

The present work has arisen with the purpose of providing a contribution to the BCI systems, through the field of Artificial Intelligence; thus, a brain-computer interface (BCI) has been developed, aimed at the management of a computer and mainly oriented to people who have reduced mobility in their upper limbs. Since a BCI system is characterized by making use of the user’s brain signals by relating certain brain patterns, the use of eye winks was established as patterns or control signal strategies. By virtue of this, the Emotiv Insight headset was selected as the EEG signal capture device. The treatment of the signals obtained was carried out in the feature extraction stage, in which aspects of time series and time windows were considered for the process of signal labeling and event determination. For the signal classification process, a MLP neural network was used as the algorithm for classifying the data set of features, with the classifier achieving a classification accuracy of 91.53 % during the training process. This allowed to generate a classification model, in order to classify in real time the signals captured by the headset on three possible actions of the mouse events: left click, right click and none. To control the mouse cursor through the user’s head movements, the headband accelerometers and the Kalman Filter were used as a tool to minimize noise and estimate the real position of the mouse cursor. Finally, tests were carried out to validate the system, including healthy people and people with some type of physical limitation in their upper limbs, achieving an accuracy of 88.06 %.