Interested Object Detection based on Gaze using Low-cost Remote Eye Tracker

Abstract

Eye tracking technology provides many advantages in human-computer interaction and human-robot interaction research. However, most studies in human-computer interaction are based on prohibitively expensive professional eye-trackers, which cannot be affordable for daily life applications. Though consumer-level low-cost eye trackers have appeared in market in recent years, they can only output raw gaze recordings and lack the high-level measurements/features for further analyzing the user's implicit visual intention in an image. To this end, a gaze-based intention detection approach using a low cost remote eye tracker is proposed. In this study, we form the gaze point into clusters and extract gaze-based features from the clusters. Then a support vector machine (SVM) classifier is used for detecting intentional gaze clusters. According to our experiment results obtained from 15 subjects, the accuracy for classifying the gaze clusters is 97.85%. Besides, the performance of our approach for detecting user's interested objects from the visual image of the real-life scene is comparable with the state-of-the-art method that employs an expensive professional eye tracker. The overall results suggest that our approach based on a low-cost remote eye tracker is applicable for detecting user's interested object.