EOG Signal Classification Based on Blink-to-Speak Language

Given the need for a universal communication method for people who cannot communicate using oral language (as speech disability, limited or no movement in the upper limbs avoiding the use of sign language, among others), the electrooculogram (EOG) is proposed as a tool for recording eye patterns. Therefore, a total of 10 patterns derived from the Blink-to-speak language, which contains essential needs, were carefully chosen to be recorded by 20 participants. From these recordings, various statistical and signal measurements were extracted as distinctive features. Subsequently, a comprehensive database was constructed using these features in order to train a machine-learning algorithm, specifically a decision tree model. The primary objective of this model was to accurately detect and predict the specific eye-movement patterns that corresponded to a given word or communicative concept, all while considering the underlying intention behind each movement. An 83.4% of accuracy were achieved for this multiclass task, the patterns that achieved 100% correct predictions were “I want to sleep” and “danger”, while the patterns with lower performance were “yes” and “toilet”. Hence, our proposal represent a viable communication alternative for those who cannot communicate due to speech and limb movement limitations.