Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : May 10, 2017 - May 13, 2017
In order to improve the quality of life, it is necessary to record and review daily-life activities by oneself. In this research, we focus on review of eating habits because they are closely related to health. For this purpose, we propose a system to summarize eating scenes by detecting eating motions. The user reviews her/his eating habits by watching the eating scenes on the summarized, short video. The proposed system aggressively uses a personalized detector for each case, which reduces the complexity of the classifier and enables to train the classifier only from the first eating scene. In the experiments, we verify that the proposed system detects 86 percent of eating motions from the eating scenes of five subjects and the duration of the summarized video is only 11 percent of the original video.
