A robust approach for constructing a graph representation of articulated and tubular-like objects from 3D scattered data

Werghi, Naoufel

doi:10.1016/j.patrec.2005.10.002

Cited by 3 publications

(1 citation statement)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Graph-theoretic analysis of 3D objects leads to a stable and dependable performance for 3D shape analysis. Reeb graph, in particular, provides a natural representation that is suitable for object surface representation due to its onedimensional graph structure and invariance to both global and local transformations [16,1,3]. Given an unorganized cloud of 3D points, identification of human body parts has been attempted in several works.…”

Section: Proposed Algorithmmentioning

confidence: 99%

Segmentation of 3D Articulated Components by Slice-Based Vertex-Weighted Reeb Graph

Karmakar

Bhowmick

Biswas

2014

Lecture Notes in Computer Science

View full text Add to dashboard Cite

A fast and efficient algorithm for segmentation of the articulated components of 3D objects is proposed. The algorithm is marked by several novel features, such as DCEL-based fast orthogonal slicing, weighted Reeb graph with slice areas as vertex weights, and graph cut by exponential averaging. Each of the three sets of orthogonal slices obtained from the object is represented by a vertex-weighted Reeb graph of low complexity, as the slicing is done with an appropriate grid resolution. Each linear subgraph in a Reeb graph is traversed from its leaf node up to an articulation node or up to a node whose weight exceeds a dynamically-set threshold, based on exponential averaging of the predecessor weights in the concerned subgraph. The nodes visited in each linear subgraph are marked by a unique component number, thereby helping the inverse mapping for marking the articulated regions during final segmentation. Theoretical analysis shows that the algorithm runs faster for objects with smaller surface area and for larger grid resolutions. The algorithm is stable, invariant to rotation, and leads to natural segmentation, as evidenced by experimentation with a varied dataset.

show abstract

Section: Proposed Algorithmmentioning

confidence: 99%