A Deep Learning Method to Estimate 3D Point of Regard by Joint Head and Eye Information

The development of systems that can characterize the state of the human is now important for many applications. In particular, as an indicator of attention and interest, the human gaze is an important cue in people behaviors, personality, intentions, and activities. Gaze also play a crucial role in the communication process. However, in spite of great advances during last three decades, current gaze estimation methods cannot addresses required conditions in this field, e.g. user head movements and minimum user calibration. There have been some works to resolve such problems but those methods lack good precision. In this work, we have used a method for appearance-based gaze estimation using convolutional neural networks, which is multimodal. This method in our implemented setting significantly outperforms state-of-the-art methods.