In today's educational institutions, PCs and tablets have been introduced, and online lectures and e-learning platforms are being utilized. In these learning environments, live video streaming and recorded videos replace face-to-face classes, while exercises and exams are conducted online. However, some users (students) do other works and do not watch the playing video. In this paper, we introduce a method for finding representative frames of students’ behaviors from recorded video, aiming to clearly visualize their actions during online learning.