Eye Movement Events Detection with KNN-GA and Prior Knowledge

Abstract

Aiming at the problems of difficulty in threshold adjustment and low detection accuracy and efficiency, an eye movement event detection method combining K-Nearest Neighbor and genetic algorithm (KNN-GA) and prior knowledge is proposed. Firstly, design the absolute amplitude feature of eye movement to describe the eye movement event characteristics of PSOs, and then the KNN is used to pre-detect eye movement events based on the minimum feature subset generated by genetic algorithm; After that, screening rules based on prior knowledge are ted to further adjust and optimize the pre-detection results. Experimental results show that this algorithm avoids threshold adjustment, and its execution efficiency is equivalent to simple IVT and NH, meanwhile the detection accuracy of fixation, saccade, and PSOs are increased by at least 3.4%, 4.7%, and 12.7%, respectively, and the detection performance is robust under different noise levels.