University of Twente Student Theses

Login

Extracting regularities from point clouds of indoor scenes

Zhu, Yunmeng (2017) Extracting regularities from point clouds of indoor scenes.

[img] PDF
5MB
Abstract:Nowadays, anincreasing number of people are interested in the build-up indoor environment as itis animportant place to optimize our quality of life in social, entertainment and economic activities. Furthermore, Indoor Mobile Laser Scanning(IMLS)makes indoor data acquisition and 3D model generation comestrue. However, the data collected by IMLSis a large set of dense point clouds, i.e. a list of point coordinates that lacks structural information in its raw stage. The data may have data gaps, or outliers which hinder the way of understanding structure and reconstruction of buildings. One way to solvethis problem is to extract regularities in the man-made indoor environment. This can help to understand and reconstruct a fundamental frame of the indoor scenes.This research was designed to detect two main characteristics that exist in an indoor environment. One is symmetry and the other is repetitive patterns. The polygon symmetry detection has 4 main steps.1.Segmentation (surfacegrowing) for raw point clouds;2.Extract segments` boundary polygonsand deal with intersection;3.Symmetry centroid selection and symmetry line generation;4.Over flip polygonsto detect symmetry and identify the symmetric parts. This process has been tested on both 2D floor plan and 3D laser point clouds. The surface growing segmentation method was applied to extract the planar surfaces from unstructured laser point cloud data. The boundary was extracted by 2D alpha shape and the boundary points` x, y coordinates were kept. To detect symmetry, a draft symmetry centroid was selected first and then refined by shifting around neighbouring regions. The remaining symmetric polygons are visualized in different colours for better recognition. Repetitive structuresin the indoor environmenthasidentifiedin this researchinclude: room width and length, wall thickness, roomconcave corners, etc.Those repetitive structures can be combined and applied to detect the similar rooms. Two approaches introducedto identify the similar rooms.1.Identification of the similar roomsby weight. Assigning a stochastic weight for each feature and selectbest results depends on different weights combination.It produces several good results.2.Hierarchical identification of the similar rooms. Rooms are grouped by the hierarchical features.The accuracy of this method canreach above 95%in some cases.Theseregularities couldprobablybeapplied in data compression, registration of point clouds, or deriving a CAD or BIM model.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:https://purl.utwente.nl/essays/85862
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page