An Implementation of SVM-Based Gaze Recognition System Using Advanced Eye Region Detection

Abstract

There have been many recent studies on gaze recognition in the field of Human-Computer Interaction (HCI). Gaze recognition and other biomedical signals will be a very natural and intuitive part of Human-Computer Interaction. In studies on gaze recognition, identifying the user is the most applicable task, and it has had a lot of attention from many different studies. Most existing research on gaze recognition has problems with universal use because the process requires a head-mounted infrared Light Emitting Diode (LED) and a camera, both expensive pieces of equipment. Cheaper alternatives like webcams have the disadvantage of poor recognition performance. This paper proposes and implements the Support Vector Machine-based (SVM) gaze recognition system using one webcam and an advanced eye region detection method. In this paper, we detected the face and eye regions using Haar-like features and the AdaBoost learning algorithm. Then, we used a Gabor filter and binarization for advanced eye region detection. We implemented a Principal Component Analysis (PCA) and Difference Image Entropy-based (DIE) gaze recognition system for the performance evaluation of the proposed system. In the experimental results, the proposed system shows 97.81% recognition of 4 directions, 92.97% recognition of 9 directions, demonstrating its effectiveness.