1A1-D01 移動ロボットのための物体認識と自己位置推定における視覚情報処理

概要

論文の詳細を見る
When a robot manipulates some objects, the object should be specified by the robot vision. This paper describes a novel object recognition approach which combines a method for generic visual categorization with three- dimensional segmentation method. The approach is as follows; first, the object area in the image is cut out by clustering 3D map of features obtained by SLAM. Next, by using the generic visual categorization method, "Bag-of-Keypoints," to the area, robot can recognize what it is and where it is in the image. Our approach contributes to the object recognition that can correspond to the diversity of objects and the image segmentation that uses 3D positions of features which compose the object. In this paper, we discuss the effectiveness of our approach through some tableware detection experiments.
一般社団法人日本機械学会の論文
2009-05-25