Deep Learning based Eye Gaze Tracking for Automotive Applications: An Auto-Keras Approach

We propose a deep neural network-based gaze sensing method in which the design of the neural architecture is performed automatically, through a network architecture search algorithm called Auto-Keras. First, the neural model is generated using the Columbia Gaze Data Set. Then, the performance of the solution is estimated on an online scenario and proves the generalization ability of our model. In comparison to a geometrical approach, which uses dlib facial landmarks, filtering and morphological operators for gaze estimation, the proposed method provides superior results and certain advantages.