Hierarchical 3D line restoration and 3D planar reconstruction in structured environments구조적인 환경에서 계층적인 방법을 통한 3차원 직선과 평면 복원

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 724
  • Download : 0
For structured environments, recent studies have represented scenes with the high-level geometric primitives, such as line and plane, rather than sparse point-based representation such as Structure-from-Motion. Most of planar reconstruction approaches extract 2D line segments and match them, then reconstruct 3D structure using a triangulation method. However, these line-based approaches are not robust against a small matching error due to the complex line extraction and matching steps. In this thesis, we propose a robust line-based 3D planar reconstruction method which does not require any line matching or triangulation steps. We first present a 3D line hypothesis generation approach using point cloud and 2D line segments. We exploit a PCA-based line hypothesis generation method to enhance the robustness against the noise of 2D line. Through systematic validations on synthetic data, we demonstrate that our generation method outperforms a conventional triangulation method at least 4 times in terms of angular accuracy. A 3D reconstruction method based on a hierarchical framework subsequently restores the 3D lines of structured environments. The restoration of noisy 3D lines still remains as a challenging problem because it is difficult to define a suitable similarity measure discriminative to other lines. The robustness of our proposed approach is due to the fact that most structured scenes consist of sets of parallel 3D lines with the same angular proximity, which provides a hierarchical similarity measure for structured 3D lines. We hierarchically cluster 3D lines based on angular and distance similarities, then the center of each cluster becomes the final 3D line restoration. The framework also makes the clustered 3D lines align along the associated angular directions. In the quantitative experiment, the hierarchical method shows 10 times better performance than the conventional cylinder model. We then present a general 3D planar reconstruction framework b...
Advisors
Kweon, In-Soresearcher권인소
Description
한국과학기술원 : 로봇공학학제전공,
Publisher
한국과학기술원
Issue Date
2014
Identifier
568860/325007  / 020123690
Language
eng
Description

학위논문(석사) - 한국과학기술원 : 로봇공학학제전공, 2014.2, [ vii, 40 p. ]

Keywords

PCA; 에너지 최소화; 계층적 군집화; 구조적인 환경; 3차원 복원; 주성분 분석; 3D Reconstruction; Structured environment; Hierarchical clustering; Energy minimization

URI
http://hdl.handle.net/10203/197202
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=568860&flag=dissertation
Appears in Collection
RE-Theses_Master(석사논문)
Files in This Item
There are no files associated with this item.

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0