A Comparative Study of Various Convolutional Neural Network Architectures for Eye Tracking System

Abstract

Eye tracking system is a technology where it can detect, trace, and analyze the whole eyes movements. Eye tracking system has been used in many fields and sectors so the number of studies for this system has been increased significantly. Eye tracking method has been switched to machine learning. One of the most used algorithms for this system is Convolutional Neural Network (CNN) which has a lot of architectures, such as LeNet-5, AlexNet, VGG16, GazeNet, and ResNet-18. This paper aims to find the most appropriate Convolutional Neural Network’s Architecture for building an eye tracking system by comparing chosen architectures that have been mentioned above. The study showed that ResNet-18 has the highest result as it gets 90.53% accuracy while the other results are lower than it.