Quick Models for Saccade Amplitude Prediction

This paper presents a new saccade amplitude prediction model. The model is based on a Kalman filter and regression analysis. The aim of the model is to predict a saccade’s am-plitude extremely quickly, i.e., within two eye position samples at the onset of a saccade. Specifically, the paper explores saccade amplitude prediction considering one or two sam-ples at the onset of a saccade. The models’ prediction performance was tested with 35 subjects. The amplitude accuracy results yielded approximately 5.26° prediction error, while the error for direction prediction was 5.3% for the first sample model and 1.5% for the two samples model. The practical use of the proposed model lays in the area of real-time gaze-contingent compression and extreme eye-gaze aware interaction applications. The paper provides theoretical evaluation of the benefits of saccade amplitude prediction to the gaze-contingent multimedia compression, estimating a 21% improvement in com-pression for short network delays