A high performance asynchronous EOG speller system

Abstract

In this study, an asynchronous region based electrooculogram (EOG) speller system has been proposed with fourty two targets consisting of alphabets, numbers and special characters. The total set of targets were arranged into six groups and each group has seven targets. The target groups were displayed in the form of a hexagon, three target groups on each side (right and left) of the display. The proposed system uses single blink or wink eye movements for selecting the desired target. It adopts two step recognition approach, 1) group selection and 2) target identification from the selected group. A pair of opposite target groups were highlighted at the same time in a random order, like an oddball paradigm. The paradigm randomly highlights all target groups once in a single trial. The subject should blink or wink for target group selection in synchrony with the highlights. In a similar manner the desired target is identified from the selected target group. A multi thresholding algorithm is used for eye movement detection. The online and offline experiments were conducted on ten healthy subjects for validating the proposed speller system. An average online classification accuracy of 98.482% is obtained with the information transfer rate (ITR) of 126.656 bits/min.