[1L5-OS-18b] 生体信号を活用した医療・ヘルスケアAI

[1L5-OS-18b-01] A survey on the image and video processing for dogs and a preliminary study on its applicability

〇Takatomi Kubo^1,2, Yurina Wada^1,3, Yuki Maruno³, Mimi Okumura³, Yuki Ura¹, Haru Yoshimura², Miho Nagasawa² (1. Nara Institute of Science and Technology, 2. Azabu University, 3. Kyoto Women's University)

Keywords:dog, image processing, deep learning, transfer learning, pre-training

Dogs are a very familiar animal to us humans, but research on their cognition, behavior, and emotions is still actively being conducted.
Among them, attempts at quantitative evaluation of their behavior have shown progress in recent years with the development of advanced machine learning methods.
Input information used includes images, videos, physiological signals such as electrocardiograms, and inertial sensor signals.
In this paper, we introduce the results of a survey on methods that use images and videos, which are frequently used because they are easy to measure.
Image/video processing for dogs covers a wide range, including detection of the dog body region, pose estimation, and emotion recognition.
In this paper, we give an overview of these processing methods and report our preliminary test of the applicability of a part of these methods.