[2C1-GS-12] Human interface, education aid: Estimating states in education (1)

[2C1-GS-12-03] Evaluation of Concentration by Human Pose Estimation from Images of Classroom

〇Tomoya Fujishiro¹, Yukimi Asari², Masahiro Miyata¹, Tetsuji Yamada¹, Takashi Omori¹ (1. Tamagawa University, 2. TOHO SYSTEMS SERVICE CO.,LTD.)

Keywords:Interest estimation, Educational support, Image analysis

We have proposed a method for estimating interest from the position and orientation of children in the image of elementary school classes to support elementary education. To automate this, we have extracted the 2D skeleton information of the child from the video using AI technology and estimated the 3D pose from it. However, there were fluctuations in extracting the skeleton, and there was a problem with the accuracy. Therefore, in this presentation, the fluctuation in skeleton extraction is corrected and removed, and the posture estimation is stabilized. Then, we aim at estimating the degree of concentration of each child in the class by classifying the change of the posture in a time series and evaluating the change of the posture of the child per unit time.