A Hybrid Model for Pupil Detection Using FPGA and CPU by Iterative Circle Fitting

Abstract

Pupil detection is a critical requirement in security applications, ocular characterization, and automated automotive systems. An increasing number of applications are being developed that use the pupil response as a measurement of cognitive function and physiological stress. This paper proposes a novel approach to pupil detection that integrates an image processing system into the Field Programmable Gate Array (FPGA) hardware of a micro controller. The FPGA is programmed to segment the pupil contour based on the pixel intensities and the CPU is used to run a circle fitting model to predict the coordinates of the pupil. This model is evaluated with a private data set and a public data set, and it outperforms the stat-of-the-art models achieving a pupil segmentation accuracy of 0.9919 and a precision of 0.9930. This model is appropriate for deployment in real-time settings for several security and surveillance applications.