Performance evaluation of EOG based HCI application using power spectral density based features

Interestingly, in recent years active research is going on in the area of signal processing using Electrooculography signals (EOG). The reason behind this lies in the fact that the eye signal remains intact even though most of the body parts fail to function normally. In this context, the eye movements can be used to interact with the computers, hence leading to the need for human computer interface (HCI) development. This paper considers four basic tasks like blinking, reading, eyes closed and open for task based classification. Here, horizontal and vertical EOG signals derived from a group of healthy volunteers are considered. Power spectral density (psd) based features are extracted and are used to classify the four basic tasks. As a pilot study, k nearest neighbor (kNN classifier) and linear discriminant analysis (LDA) has been used. The results indicate that the highest classification accuracy is obtained using Yule Walker's estimation for vertical EOG. Also, performance of the kNN using three nearest neighbors stands out the best possibility with the classification accuracy of 81.87%. Also, the results rule out the possibility of using linear discriminant analysis for the task based classification using EOG signals.