3D shape reconstruction with a monocular camera by employing geometric primitives

2018 Fiscal Year Final Research Report

• Project/Area Number: 16K16084
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Perceptual information processing
• Research Institution: National Institute of Advanced Industrial Science and Technology (2018); Toyohashi University of Technology (2016-2017)
• Principal Investigator: Oishi Shuji 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 研究員 (30759618)
• Project Period (FY): 2016-04-01 – 2019-03-31
• Keywords: 3次元復元 / Visual SLAM / Model fitting

Outline of Final Research Achievements

Accurate 3D shape measurement technology based on limited sensor information obtained from a specific viewpoint is important to grab a structure of the scene. In this research, we addressed dense 3D reconstruction using a monocular camera. Specifically, we developed a Pixel-wise depth estimation method and a new Visual SLAM that maintains an enormous amount of feature points to generate highly-detailed 3D models with a monocular camera. In addition, in order to achieve denser shape recovery, we developed a model fitting method where a reference model (a geometric primitive) is aligned toward a partial shape observation, which leads to continuous 3D model reconstruction from a discrete 3D point cloud.

Free Research Field

ロボットビジョン

Academic Significance and Societal Importance of the Research Achievements

移動ロボットのような実世界で稼働するシステムでは，正確な3次元環境計測が課題となる．人間と同様，移動ロボットも現在地点からの限定的な観測のみ利用可能であり，その断片的な情報の蓄積から高精度・高密度な形状情報を得ることで安全な経路計画や物体操作が実現できる．
本研究は，一般に用いられる単眼カメラを用いた3次元形状復元手法を開発するもので，特殊なセンサを必要としないため多様なシステムへ適用が可能である．