Gaussian Distribution based Urban Scene Understanding in 3D point clouds for Robotic Applications = 로봇에 응용을 위한 3차원 점군에서의 정규 분포 기반 도시환경 물체 인식기법

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 562
  • Download : 0
Urban scene understanding is the ability to categorize ambient objects into several classes and it plays an important role in various urban robotic missions, such as surveillance, rescue, and SLAM. In spite of many previous works in urban scene understanding, it is still hard to apply it to urban robotic missions because point clouds scanned in urban environments are complex and massive. In this thesis, we address the difficulties and propose urban scene understanding algorithms which solve the difficulties. To achieve this, the proposed algorithms in this thesis were designed based on Gaussian distribu-tions. Gaussian distribution is a useful tool to simply represent point clouds in terms of mean and covari-ance, instead of directly dealing with complex and massive point clouds. By doing so, we tried to derive useful properties for robotics applications from Gaussian distributions. In this thesis, we propose two kinds of urban scene understanding algorithms based on Gaussian distributions. The two proposed algorithms have respective advantages for robotic applications. For the first algorithm, we propose an urban scene understanding algorithms based on Normal Dis-tribution Transform (NDT) grids for robots. In 3D NDT grids, point clouds are stored in the form of mean and covariance. The advantage of using 3D NDT grids is that massive 3D point clouds can be represented by simple the parameters in 3D grids, i.e mean and covariance. Moreover, NDT grids support incremental update of point clouds without increasing memory size. This strength is useful for storing dense or huge amounts of point clouds. Based on NDT grids, we design Geometric-Featured Voxel (GFV) to represent urban structures as a voxel model. The proposed method consists of three steps: GFV generation, segmentation, and classification. Experimental results prove that the proposed method is suitable for robotic applications in terms of memory requirement, computation time, and even classification acc...
Advisors
Chung, Myung-Jinresearcher정명진
Description
한국과학기술원 : 로봇공학학제전공,
Publisher
한국과학기술원
Issue Date
2014
Identifier
591748/325007  / 020105196
Language
eng
Description

학위논문(박사) - 한국과학기술원 : 로봇공학학제전공, 2014.8, [ vii, 90 p. ]

Keywords

Scene understanding; 가우시안 분포; 도시환경 구조물 분류; 3차원 점군; 레이저스캐너; 도시환경; Urban Environment; LIDAR; Point Cloud; Gaussian Distribution

URI
http://hdl.handle.net/10203/197131
Link
http://library.kaist.ac.kr/search/detail/view.do?bibCtrlNo=591748&flag=dissertation
Appears in Collection
RE-Theses_Ph.D.(박사논문)
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