Controle de uma cadeira de rodas utilizando potencial evocado visual

Abstract

This work aims at developing a steady state visual evoked potential (SSVEP) braincomputer interface (BCI). For this purpose, the work was divided in two steps: i) performance investigation of two Objective Response Detection (ORD) techniques, the Magnitude Squared Coherence (MSC) and its multivariate version, the Multiple Coherence (MC), for SSVEP detection in two different lighting conditions: scotopic (low luminance) and photopic (daylight compatible luminance); ii) implementation of a SSVEP-BCI for wheelchair control. For luminance influence evaluation, EEG signals were collected from ten volunteers. stimulated with frequencies of 6, 7, 8, 9, 10 and 11 Hz (random presentation order). MSC was applied to parietal ([P7],[P8]) and occipital ([O1,O2]) derivations and MC to the combinations [O1-O2], [O1-O2-P7], [O1-O2-P7] and [O1-O2-P7-P8]. The developed BCI has 3 commands, left, right and forward, associated, respectively, to the frequencies 7.75; 8 and 10 Hz. For feature extraction, three ORD techniques were used, MC, MSC and Spectral F Test (SFT), being the first one employed for online experiments, in which 34 volunteers sat on a wheelchair and received random instructions to focus their attention on a given direction. The EEG recording was performed using a 14-channel wireless system. In order to evaluate the BCI performance using the different techniques, seven parameters were employed: number of detections, mean time between detections, hit rate, information transfer rate (ITR) and false positives for each direction. The detection rates using MSC showed a significant decrease at photopic condition compared to the scotopic one for all frequencies and derivations. The MC application to [O1-O2-P7-P8] allowed overcoming this limitation, leading to detection rates higher than 80% for all frequencies (except 9 Hz). The best performance was achieved with MC that reached, in average, the highest number of detections (241), hit rate (85%) and ITR (24 bits/min). On the other hand, SFT presented the worst results for these three parameters.