Change search
ReferencesLink to record
Permanent link

Direct link
Robot vision: experimental segmentation of thin structures in 3D laser data
2000 (English)Independent thesis Advanced level (professional degree), 20 credits / 30 HE creditsStudent thesis
Abstract [en]

An experimental low-level segmentation algorithm working with 3D laser data has been developed. Research around scene analysis and segmentation mostly deal with classifying big and/or "well defined" objects with smooth surfaces. As a consequence, those methods are often noise insensitive. So they are not primarily aimed at detecting small and "ill-defined" objects that may be interpreted as noise. This thesis makes an attempt at identifying thin and straight objects. The input is a Delaunay triangulation of a spherical projection of 3D datapoints generated by a range scanning laser. The output is a set of clusters that have datapoints homogeneously distributed within themselves with respect to some measure. The choice of measure is crucial to the result. The clustering is done by measuring the local homogenity of closeness/distribution between datapoints connected by the graph-edges in a nearest neighbour tree. Those edges not being homogeneous in their local neighbourhood are eliminated and the remaining edges define clusters of datapoints. The clusters are then interpreted as pipes by a RANSAC-type algorithm guided by the medial axes of the clusters. Experimental results for elongated objects are presented.

Place, publisher, year, edition, pages
2000.
Keyword [en]
Technology, segmentation, clustering, surface filtering, range data, range measuring laser
Keyword [sv]
Teknik
Identifiers
URN: urn:nbn:se:ltu:diva-47782ISRN: LTU-EX--00/214--SELocal ID: 54b1c7ac-b0f0-45f7-b24e-a766505fe94bOAI: oai:DiVA.org:ltu-47782DiVA: diva2:1021111
Subject / course
Student thesis, at least 30 credits
Educational program
Civil Engineering programmes 1997-2000, master's level
Examiners
Note
Validerat; 20101217 (root)Available from: 2016-10-04 Created: 2016-10-04Bibliographically approved

Open Access in DiVA

No full text

Search outside of DiVA

GoogleGoogle Scholar

Total: 1 hits
ReferencesLink to record
Permanent link

Direct link