SSVEP-based BCI with visual stimuli from LCD screen applied for wheelchair control: offline and online investigations

• Autor(a) principal: Zambalde, Ellen Pereira
• Data de Publicação: 2018
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Institucional da UFU
• Texto Completo: https://repositorio.ufu.br/handle/123456789/22316; http://dx.doi.org/10.14393/ufu.di.2018.1166
• topic: CNPQ::ENGENHARIAS::ENGENHARIA BIOMEDICA::BIOENGENHARIA::PROCESSAMENTO DE SINAIS BIOLOGICOS; SSVEP based BCI; LCD screen; Target location; Window length; SVM; ICM baseada em PEV-RP; Controle de cadeira de rodas; Monitor LCD; Localização de alvos; Tamanho do janelamento; Wheelchair control; Engenharia biomédica; Cadeiras de rodas - Controle automático

Brain Computer Interfaces (BCIs) have been shown as a promising technology in the current years, especially for those wheelchair users affected by motor injuries or diseases. Steady-State Visual Evoked Potentials (SSVEP)-based BCI systems are widely used for many applications, such as keyboard and robot control, because of its short response time and ease of use. SSVEP-based BCIs use brain responses to any visual stimulus flickering at a specific frequency as input command to an external application or device. Although some SSVEP-based BCI applied to wheelchair control have shown promising results, there are specific features of the system that should be analyzed and discussed aiming to increase classification accuracy. This dissertation aims to develop and investigate the performance a SSVEP-based BCI system using LCD monitor as visual stimulator applied for wheelchair control. Two experiments were performed (offline and online). Through the offline experiment, performed by 9 participants, it was possible to identify EEG optimal channels, adequate window length for signal processing, and target location on the LCD screen. In the online experiment, 9 EEG channels (PO3, PO4, PO5, PO6, PO7, PO8, O1, O2 and Oz) were used to record the brain signal while using an interface with 5 stimuli targets (15, 12, 6.67, 8.57 e 10 Hz) placed on the top, bottom, right, left and center of the screen. Four participants were positioned in front the LCD screen where the interface was executed. The interface was developed in Python language and, after collecting the signal, it was responsible to perform filtering, windowing, feature extraction (FFT), and classification (SVM). The results showed a good performance while using a 3 seconds time-window with 250 ms of overlap. Accuracy rates from online experiments were high, which allowed the control of a powered wheelchair. This system can provide independency and increase quality of life of wheelchair users. Future works involve adaptations and improvements of the system to increase efficiency and provide more reliability to the user.