Prediction of Oblique Saccade Trajectories Using Learned Velocity Profile Parameter Mappings

Abstract

This manuscript proposes and validates two techniques for predicting the trajectory of oblique saccades using a Gaussian velocity profile model. Profile parameters and event duration are estimated at the onset of each saccade using support vector machine regression models. The proposed techniques are evaluated using a set of 47,652 saccades with a mean amplitude of 12.25 degrees of the visual angle gathered from 322 subjects during a random saccade task. Numerous performance metrics are evaluated for predictions made at various fractions of the saccade duration. An average landing point estimation error of less than three degrees of the visual angle is obtained for predictions formed at 30% of the saccade duration.