Extraction de potentiels évoqués P300 pour les interfaces cerveau-machine

Brain-computer interfaces (BCI) enable non-muscular powerful channel for communicating thanks to direct communication between the user’s brain and a computer. In this article, the presented BCI concerns the P300 speller which enables people to write a text on a computer by estimating P300 evoked potentials. In this work, an unsupervised algorithm is introduced for signal subspace estimation: the raw EEG are thus enhanced by projection on the estimated subspace. The detection of P300 potentials is performed by a linear support vector machine (SVM). The proposed method is finally shown to be efficient since it increases the number of spelt characters per minute.