Exploitation of EEG-extracted eye movement for a hybrid SSVEP home automation system

Abstract

Brain-computer interface (BCI) systems allow a direct communication between a user and a computer using only brain activity. BCIs convert electrical neurosignals, recorded through electroencephalography (EEG), into actual commands to operate a software application or a device. Among the various neurophysiological phenomena that can be used to drive BCI systems, steady state visual evoked potentials (SSVEPs) have demonstrated the highest performance for BCI systems. The flickering stimuli required by these systems tend to be annoying for the user and the accuracy of SSVEP-based BCIs tends to also decrease as the number of flickering stimuli increases. To address these issues, the project aims to extract EEG potentials related to eye movements to estimate the point of gaze of the user when looking at the stimuli and hence obtain a broad idea of the stimulus the user is focusing on. This information will be used such that stimuli which are far from the user’s point of gaze can be switched off prior to the flickering of the stimuli. This reduces the number of simultaneously flickering stimuli which should improve both the annoyance factor and the classification accuracy. To test out this hypothesis, an offline study is conducted to investigate what type of eye-movement can be reliably detected from EEG and how the number of frontal channels considered and size of the training set influences the detection of eye-movements. Results have shown that a 99% accuracy is achieved when discriminating between horizontal and vertical eye-movements with three frontal channels and with 16 training trials. Subsequently a real-time hybrid BCI (hBCI) which fuses SSVEPs with EEG-based eye-movement potentials is developed. A smart home system is designed and integrated with an SSVEP-based BCI and also with the developed hBCI. A comparative analysis is conducted to evaluate the differences between the two systems and significant differences in accuracy and efficiency were found between the two. A 72.8% accuracy, an 82.4% efficiency and an ITR of 28.6bpm are achieved by the proposed hBCI whereas a 61.5% accuracy, a 74% efficiency and an ITR of 27.6bpm are achieved by the SSVEP-based BCI.