Proposal of a Person-Tracking Audio Input-Output System Using Directional Microphone and Directional Speaker
Masaki Mitsuyama, Seiichi Serikawa · 2023
Care robots are becoming more and more important. There are various doors and doorways in homes, but it is difficult for a robot to move through them. Therefore, as a preliminary step in developing an autonomous mobile robot that can open and close any door, this paper aims to identify doorknobs and handles in images using machine learning. Houses have various types of doors and doorknobs. In this study, they are classified into 6 patterns. We use a Convolutional Neural Network (CNN) to classify them from images. We trained six typical doorknob patterns, 800 images were prepared for each pattern. As the result, the recognition accuracy improved to 79.9% compared to a conventional method. In future, in order for a robot to open and close any door, it will be necessary to detect 3D position of door.