Hierarchical data clustering approach for segmenting colored three-dimensional point clouds of building interiors

A range scan of a building's interior typically produces an immense cloud of colorized three-dimensional data that represents diverse surfaces ranging from simple planes to complex objects. To create a virtual reality model of the preexisting room, it is necessary to segment the data into meani...

Full description

Saved in:
Bibliographic Details
Published in:Optical Engineering 2011-07, Vol.50 (7), p.077003-077003
Main Authors: Sareen, Kuldeep K, Knopf, George K, Canas, Roberto
Format: Article
Language:English
Subjects:
Algorithms
Clouds
Clustering
Clusters
Curvature
Density
Segmentation
Three dimensional
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:A range scan of a building's interior typically produces an immense cloud of colorized three-dimensional data that represents diverse surfaces ranging from simple planes to complex objects. To create a virtual reality model of the preexisting room, it is necessary to segment the data into meaningful clusters. Unfortunately, segmentation algorithms based solely on surface curvature have difficulty in handling such diverse interior geometries, occluded boundaries, and closely placed objects with similar curvature properties. The proposed two stage hierarchical clustering algorithm overcomes many of these challenges by exploiting the registered color and spatial information simultaneously. Large planar regions are initially identified using constraints that combine color (hue) and a measure of local planarity called planar alignment factor. This stage assigns 72 to 84% of the sampled points to clusters representing flat surfaces such as walls, ceilings, or floors. The significantly reduced data points are clustered further using local surface normal and hue deviation information. A local density driven investigation distance (fixed density distance) is used for normal computation and cluster expansion. The methodology is tested on colorized range data of a typical room interior. The combined approach enabled the successful segmentation of planar and complex geometries in both dense and sparse data regions.
ISSN:0091-3286
1560-2303
1560-2303
DOI:10.1117/1.3599868