Multicore and GPU Based Pupillometry Using Parabolic and Elliptic Hough Transform

Abstract

Pupillometry is the measure of pupil size and has pertinent applications in the area of neuroscience, cognitive behavior assessments and different psychologically evoked responses. In this work, Hough transform, a high-fidelity accurate technique, is used for pupil detection and measurement. The eye images are taken from an off-theshelf webcam. During the eye segmentation process, eyelasheyelid detection is based on parabolic Hough transform whereas iris and pupil detection is done through elliptic Hough transform. The developed eye segmentation algorithm show high accuracy, however, is computationally expensive. To deal with the problem, a parallel data distribution framework is employed that uses the raw computational power of emerging multicore CPUs and many core GPUs thereby boosting the performance of proposed algorithm. A performance comparison is carried out for sequential and parallel framework of proposed algorithm. The experimental results indicate a speed-up with a factor of 1.86x and 3.56x on a Core i7 CPU and Tesla T10 GPU respectively. The proposed parallel approach can be used to accelerate pupillometry applications on multicore and GPU based high performance computing platforms. 

GPU-Based Segmented-Beat Modulation Method for Denoising Athlete Electrocardiograms During Training

“…A CPUbased approach uses single instruction single data (SISD) programming structure executing one program statement operating on one data stream at a time. Whereas, GPU architecture is characterized by single instruction multiple data (SIMD) feature [13], which allows operations on multiple data points simultaneously. GPU provides a highperformance hardware platform for faster simulations.…”