Object Classification Considering Movability Based on Shape Features with Parts Decomposition

  • NISHIDA Takaaki
    Graduate School of Systems and Information Engineering, University of Tsukuba
  • HARA Yoshitaka
    Graduate School of Systems and Information Engineering, University of Tsukuba
  • TSUBOUCHI Takashi
    Graduate School of Systems and Information Engineering, University of Tsukuba

Bibliographic Information

Other Title
  • 物体のパーツ分割と形状特徴に基づく移動可能性を考慮した種別分類
  • ブッタイ ノ パーツ ブンカツ ト ケイジョウ トクチョウ ニ モトズク イドウ カノウセイ オ コウリョ シタ シュベツ ブンルイ

Search this article

Abstract

It is important for a robot to have an ability to classify the objects in the environment. In the people living environment such as offices, for the purpose of obstacle avoidance or object operation, it is desirable for a robot to recognize objects that have potentials to move. In this paper, an algorithm is proposed to classify the objects into the movable ones to omni-direction such as chairs with casters, moved ones usually by people such as chairs without any casters, the ones with wheels restricted moving direction such as platform trucks, the ones simply with wheels, and other objects. We use a depth camera that provides point clouds of the object: 1) We extract objects from input point cloud. 2) We segment the object into part elements by comparing horizontal sectional shapes at each height. 3) We distinguish the objects whether it has wheels or not by extracting circular shape. 4) We classify shape distribution of each part elements by principal component analysis. 5) We merge part elements which is same shape distribution, to decompose objects into parts. 6) Finally we classify the objects by shape features of the parts of objects. Furthermore, we have experiments of the classification of objects in consideration of movability in the indoor environment by a robot. The robot was able to classify correctly almost all movable objects in the room. As the experimental result, the robot was able to recognize almost all movable objects in the room.

Journal

References(6)*help

See more

Details 詳細情報について

Report a problem

Back to top