A quantitative comparison of the most sophisticated EOG-based eye movement recognition techniques

Although ElectroOculoGraphic (EOG) signals have been intensively used for human-machine interfaces, none of the available eye movement recognition techniques have been objectively compared to each other. In this paper, we propose to compare two widely known techniques (the standard R. Barea (RB) and A. Bulling (AB)'s works) and a Spiking Neural Network based approach. We also suggest several potential improvements that were all assessed according to the Fl-score. Additionally, we investigate 3 different target configurations on the screen: 3×3, 3×5 and 5×5. This aims at detecting which configuration can reach the best bitrate. Finally, double blink and wink detectors are Fl-score evaluated to estimate their relevancy as a mouse click. In this 6-healthy-subject experiment, we observed that both RB and AB methods provide fairly similar results. According to the bitrate analysis while considering complexity, the 3×3 is the most suitable interface. Among the different potential enhancements, the clustering approach instead of a fixed grid leads to a much quicker learning procedure. Regarding the eye mouse click detectors, their performance should be high enough to be used in a reliable interface.