| Title |
Visual Attention Analysis System for Children using on Deep Learning Model of Gaze Estimation with Video Stabilization Structure |
| Authors |
김호영(Ho-Young Kim) ; 박윤하(Yun-Ha Park) ; 이종택(Jong Taek Lee) |
| DOI |
https://doi.org/10.5573/ieie.2023.60.4.28 |
| Keywords |
Children's visual attention analysis; Gaze estimation; Video stabilization; Object detection |
| Abstract |
This paper proposes a deep learning-based gaze estimation model (VSL2G-Net) and a system for analyzing child attention by utilizing the proposed model. In order to analyze people's attention, it is crucial to estimate their gaze information accurately. However, children in kindergarten's face-to-face classes were mostly recorded by a camera held by their teachers due to their frequent and vigorous movements. When we applied a gaze estimation deep learning model on such video footage to analyze the child's class attention, the gaze estimation accuracy was low. To solve this problem, we propose a deep learning-based gaze estimation model that applies the structure of a video stabilization model. In addition, we propose a system that analyzes the children's visual attention by determining whether the estimated gaze passes through the area to be focused in the image. The accuracy of the attention analysis system improved from an average of 0.961 to 0.967 after applying the video stabilization structure to the gaze estimation model, compared to the model without the stabilization structure. |