This paper addresses the eye gaze estimation problem in low-resolution images, using the low-cost camera in order to eliminate problems caused by infrared high-resolution imaging such as needing an expensive camera, complex setup, special light sources, and being limite
More
This paper addresses the eye gaze estimation problem in low-resolution images, using the low-cost camera in order to eliminate problems caused by infrared high-resolution imaging such as needing an expensive camera, complex setup, special light sources, and being limited in lab research environments. In the proposed method, the human face is detected with Ant Colony Optimization (ACO) algorithm, and then the Kirsch compass mask is utilized to detect the position of humans’ eyes. For iris detection, a novel strategy based on ACO algorithm, which has been rarely used before, is applied. The pupil is recognized by morphological processing. Finally, the extracted features, obtained from the radius and position of the irises of the pupils, are given to the Support Vector Machine (SVM) classifier to detect the gaze pointing. In order to receive assurance of the reliability and superiority of the newly designed ACO algorithm, some other metaheuristic algorithms such as (GA, PSO, and BBO) are implemented and evaluated. Additionally, a novel dataset, comprising 700 images gazing at seven different major orientations, is created in this research. The extensive experiments are performed on three various datasets, including Eye-Chimera with 92.55% accuracy, BIOID dataset with 96% accuracy, and the newly constructed dataset with 90.71% accuracy. The suggested method outperformed the state of the art gaze estimation methods in terms of the robustness and accuracy.
Manuscript profile
