Deep-learning based robust edge detection for point pair feature-based pose estimation with multiple edge appearance models

Diyi Liu, Shogo Arai, Fuyuki Tokuda, Yajun Xu, Jun Kinugawa, Kazuhiro Kosuge

研究成果: Conference contribution

3 被引用数 (Scopus)

抄録

To realize a robotic bin picking system, pose estimation for the objects randomly piled up in a bin is necessary. For various types of objects, many pose estimation algorithms have been proposed so far. Point Pair Featurebased Pose Estimation with Multiple Edge Appearance Models (PPF-MEAM) has been proposed for estimating the pose of industrial parts including some parts whose point clouds are defective in our previous work. Although this method shows high performance in pose estimation under a constant environment, its performance drops under the changing light conditions without tuning parameters. To overcome this problem, we propose Deep-Learning based Robust Edge Detection (DLED) for PPF-MEAM to make it robust to changes of the light. The effectiveness of DLED is proved by the edge detection experiment under different light conditions. Moreover, the pose estimation experiment proves that DLED could improve the pose estimation performance of PPF-MEAM under different light conditions.

本文言語English
ホスト出版物のタイトルIEEE International Conference on Robotics and Biomimetics, ROBIO 2019
出版社Institute of Electrical and Electronics Engineers Inc.
ページ2920-2925
ページ数6
ISBN(電子版)9781728163215
DOI
出版ステータスPublished - 2019 12月
イベント2019 IEEE International Conference on Robotics and Biomimetics, ROBIO 2019 - Dali, China
継続期間: 2019 12月 62019 12月 8

出版物シリーズ

名前IEEE International Conference on Robotics and Biomimetics, ROBIO 2019

Conference

Conference2019 IEEE International Conference on Robotics and Biomimetics, ROBIO 2019
国/地域China
CityDali
Period19/12/619/12/8

ASJC Scopus subject areas

  • 人工知能
  • コンピュータ サイエンスの応用
  • ハードウェアとアーキテクチャ
  • 機械工学
  • 制御と最適化

フィンガープリント

「Deep-learning based robust edge detection for point pair feature-based pose estimation with multiple edge appearance models」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル