Saccadic intrusion recognition intelligent algorithms based on different eye movement date classification methods

Abstract

The saccadic intrusion recognition algorithm proposed by Tokuda doesn’t take into account individual differences and also, since the accuracy is not high enough, three aspects in this regard have been vastly improved. Firstly, the algorithm is tailored to deal with three types of missing data such as low confidence, not on the screen, and time missing, which improves the data fault tolerance of the algorithm. Secondly, the improved algorithm refers to the E&K algorithm, utilizing the ratio of saccade speed to overall speed. Then, the adaptive speed threshold has been determined accurately, which improves the sensitivity of the algorithm. Finally, the algorithm adds the upper limit of the amplitude for identifying regular saccades and filters out rapid retrospectives with large amplitude. A combination of all three minor improvements boosts the overall accuracy of the algorithm. In addition, the N-back task experiment provided data, which has been processed by the improved algorithm, and aided in arriving at a conclusion consistent with the previous ones. The experimental data compares the effect of the improved algorithm and DBSCAN in identifying fixation points. Furthermore, the results demonstrate the improved algorithm being significantly better than the DBSCAN algorithm in regard to sequence and sensitivity of data points.