User-centered Ambient Assisted Living: Brain Environment Interface

This paper presents a P300-based Brain Computer Interface for the environmental control in ambient assisted living (AAL) context. The system exploits a machine-learning algorithm based on spatio-temporal characterization of the P300, which realizes the binary discrimination from a multiclass classification. The BCI architecture is made up by (i) the acquisition unit, (ii) the processing unit and (iii) the home management. The acquisition unit is realized by a wireless EEG headset collecting data from 6 electrodes. The processing unit is a hybrid PC-Raspberry Pi architecture that: (i) provides stimulation, operates a Machine Learning (ML) and (iii) realizes a real-time multidimensional classification. The ML stage is based on a custom algorithm (t-RIDE) which emphasizes the subjective P300 characteristics and trains the classifiers. The extracted features are functionally reduced and, thus, used to define proper prediction boundaries. The Raspberry-based core classifies and actuates the received commands, interfacing the environment through a dedicated module that manages the home via TCP/IP. The experimental results on a dataset of 7 subjects enable a classification with an accuracy of 84.28% ± 0.87 % in less than 10ms.