Human computer interface (HCI) system helps disabled people to interact with their surrounding world with the help of physiological signals like, electroencephalogram (EEG), electro-oculogram (EOG) and electromyogram (EMG) etc. EEG and EOG signals are widely used in many HCI applications owing to the multiple signal patterns/components of these signals. The steady state visual evoked potential (SSVEP), P300 potential, slow cortical potential and event related synchronization/desynchronization are the components of EEG signals. EOG signals have pattern or components based on the different eye movements. HCI system uses these components as a control signal for real time applications. Among various EEG components, SSVEP has been studied predominantly due to high information transfer rate (ITR) and minimal or negligible training time. The performance of SSVEP based HCI speller systems depends on the total number of targets in the keyboard layout, classification accuracy and ITR. The practical SSVEP based keyboard systems have a trade-off between the total number of targets and classification accuracy.
In this work, the SSVEP and EOG based hybrid keyboard systems have been designed and each system is validated with 10 healthy subjects. Four different speller or experimental paradigms are designed and the data (EEG and EOG) were acquired using indigenously developed data acquisition system which consists of ADS1299 IC by Texas Instruments. The classification accuracy and ITR are considered as performance metrics for evaluating the proposed systems. The proposed keyboard systems can be broadly categorized into synchronous and asynchronous systems. In synchronous systems the users/subjects are always in control state. Subjects should follow the instruction provided by the system. In asynchronous case the users/subjects can have control on the system allowing the user to change the system mode from active to idle and vice versa. The proposed works under synchronous system category are dual frequency SSVEP-EOG and two stage EOG-SSVEP system. In dual frequency SSVEP-EOG system thirty-six unique targets are designed with the dual combination of selected frequencies. The selected frequencies are integer divisions of monitor refresh rate. The visual stimuli are segregated into three groups and each group is arranged into different regions (left/middle/right) of the keyboard/speller layout for improving the target detection rate. Later, visual feedback has added to the system for improving the performance of the system. In this case, an average classification accuracy of 98.33% is obtained with the ITR of 69.21 bits/min for all the subjects. In two stage EOG-SSVEP system thirty-six targets are divided into nine groups which include alphabets, numbers and special characters. Target selection consists of two stages. Various eye movements (gaze, blinks, winks) are used for selecting the target groups and SSVEP is used to identify the target from the selected group. The average classification accuracy of two stage EOG-SSVEP system is 94.16 % with the ITR of 70.99 bits/min.
The proposed asynchronous speller systems include the region-based SSVEP-EOG speller system and the oddball EOG speller system. The region based SSVEP-EOG speller system consists of fifty target characters grouped and indexed into five flickering unique frequency stimuli. The target character is detected by using simultaneous combination of SSVEP and eye movement signal. More targets and asynchronous control are achieved by integrating EOG into the SSVEP system. The desired target will not be selected until an eye movement is performed. The average online classification accuracy of 99.47% is obtained with the ITR of 184.48 bits/min. The oddball EOG speller system consists of forty-two target characters arranged in the form of six groups on the speller layout. The subject should perform blink or wink eye movement for target selection in synchrony with the random highlights which is given by the proposed oddball paradigm. The continuous eye movements are avoided by the proposed paradigm for target selection which increases performance of the speller system and reduces fatigue. The average online classification accuracy of 98.482% is obtained with the ITR of 126.656 bits/min. In this work, the synchronous and asynchronous keyboard/speller systems are designed and validated. The performance metrics of the proposed systems are compared with the conventional speller systems. The proposed asynchronous speller systems have attained high performance metrics and offer a better user interface for real time usage.