A System for Noncontact Estimation of Cognitive Load Using Saccadic Parameters Based on a Serio-Parallel Computing Framework

This paper proposes a system which uses a three-stage serio-parallel video-oculographic framework for computing the saccadic eye parameters to indicate the amount of cognitive loading. The three stages are viz., face and eye detection, iris and eye corner localization, and finally saccadic parameter computation. Since saccades are fast movements of the eyeballs, accurate estimation of these parameters requires high frame rates of acquisition and processing. Our proposed framework meets such deadlines by accelerating the process using graphics processing units (GPUs). The first stage comprises the face and eye detection using the respective Haar classifiers followed by tracking of a region of interest using a minimum output sum of squared error filter. In the second stage, the filter parameters are transferred to the GPU, where our proposed parallel scheme is implemented. In the detected eye region, the iris candidates are ranked using a sum of dot products of normalized displacement vectors with gradient vectors. We also localize the eye corners as the reference points. The saccadic velocity and duration are obtained using this eye position signal in the third stage. Finally, the amount of cognitive loading is determined based on these parameters.