Gaze Tracking for Hands-Free Human Using Deep Reinforcement Learning Approach

Abstract

Abstract People with disabilities have new and advanced methods to communicate with the applications for virtual keyboards and other communication tools. In this paper, we utilized a novel deep reinforcement learning-based technique for determining the location of the accessible options for gaze-controlled tree-based menu selection system. A virtual English keyboard has been incorporated into the layout of the new user interface, which also includes improved requests for text modification through the gaze. The two methods used to manage the system are: 1) eye tracking for typing on the virtual keyboard and 2) eye tracking with a device for soft-switch utilizing deep reinforcement learning. Simulation results show that DRL based algorithm outperforms other baseline techniques in terms of total loss and accuracy.