Gaze Estimation by Attention Using a Two-Stream Regression Network Dikkat Tabanli Bakiş Noktasi Tahmini Için Iki Akişli Regresyon Aǧi


Karazor A., Bayar A. E., Topal C., ÇEVİKALP H.

31st IEEE Conference on Signal Processing and Communications Applications, SIU 2023, İstanbul, Turkey, 5 - 08 July 2023 identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/siu59756.2023.10223906
  • City: İstanbul
  • Country: Turkey
  • Keywords: attention mechanism, gaze estimation, humancomputer interaction
  • Istanbul Technical University Affiliated: Yes

Abstract

Determining the point of view of people is an important human-computer interaction problem that has been studied for a long time. This subject, which has many applications, is used in different fields such as marketing, automotive, medical, games and entertainment. In this study, we propose a remote eye tracking method that makes gaze estimation using convolutional neural network based on regression. The proposed method uses a two-stream deep learning architecture that utilizes eye images and iris segmentation masks obtained through segmentation neural network. The architecture employed selective attention-based mechanisms to enhance its performance. Experimental results demonstrate that the attention-based two-stream architecture outperforms both single-stream deep learning architectures and architectures without attention mechanisms.