Detection and Cognitive Assessment of Pupil Dilation using High Performance Computing

Abstract

Pupil dilations are sensitive to human affective responses. The assessment of pupillary responses has found multi-purpose applications such as lie detection in law enforcement agencies, providing an innovative solution for human-computer interaction, minimizing dangers in air traffic displays, communicating with people suffering from ailments such as autism, developing interactive video games, eye gaze correction for videoconferencing, marketing and consumer research. In this work, the detection and assessment of pupil diameter is done by developing a low-cost eye tracking system. The scope of the system development is to accurately detect pupillary responses over a period of time i.e. subtle changes in pupil size that indicate cognitive load. The features of the proposed framework are eye and pupil-localization, pupil area and diameter calculation in a recorded video sequence and investigation of pupil size variation during and after auditory stimulation. In this case human eye images were acquired by a webcam and processed offline. The hardware includes a micro-lens webcam and head mounted structure with a USB data transfer capability to computer. The framework for the algorithm development includes iris and pupil localization, pupil area and diameter extraction and calibration that are implemented in MATLAB. Subsequently, the algorithm is implemented on video sequences from the

webcam. Pattern Recognition techniques are used to transform the images from RGB to binary and Hough Transform is used to detect pupil area from the rest of the eye image. Moreover, a high performance implementation of the proposed algorithm is done using MATLAB® parallel processing toolbox and Graphical Processor Unit (GPU). A comparison of speedups achieved by implementing the above techniques is analyzed. Finally, experiments are conducted on two subjects to measure human cognition and the algorithmic efficiency of eye pupil location and variation.