Directional Binary Pattern (DBP): A novel object representation approach for gaze estimation

This paper presents a novel object descriptor, Directional Binary Pattern (DBP), for robust gaze estimation. In DBP, the local directional derivations are proposed to encode the binary patterns in the given orientations. As an object descriptor, DBP has many advantages: noise restrain, robustness to illumination, contrast enhancement on the boundary, and local texture encoded as the directional information. DBP features of eyes are finally fed into Support Vector Regression (SVR) to match the gaze mapping function, which is then used to predict the gaze direction with respect to the camera coordinate system. In our experiments, an eye gaze dataset includes 4089 training samples of 11 persons. Experimental results show that our method can achieve an accuracy of less than 2°.