CIS Seminar: David O'Gwynn, 10/20

"Towards Semantic Alignment of Surface Mesh Shapes"

David O'Gwynn
University of Alabama at Birmingham

3:00 p.m., Wednesday, October 20, 2010
235 Weir Hall

Given the rapid increase in the variety of sources of 3D surface mesh data, there is a need for techniques that can orient this mesh data consistently. Many mesh alignment techniques exist, but they do not comprehend the semantic aspects of shape. That is, they treat meshes as a collection of surface samples, instead of as a collection of semantic parts. This makes them sensitive to variations in sampling technique and model pose. To alleviate this, we will discuss the concept of semantic alignment, which is the alignment of mesh-encoded shapes based on their decomposition into semantic parts, and introduce an algorithm for such. Preliminary tests show it to be robust to different sampling methods and minor variations in pose. We will first present the problem of mesh alignment and the limitations of current solutions. We will then present a system for encoding 0D mesh features into a 1D part-aware feature graph. Finally, we will demonstrate how this feature graph can be used to perform part-aware alignment of mesh shapes.

David O'Gwynn obtained his M.S. in Computational Engineering from Mississippi State University and is currently a PhD candidate in Computer and Information Sciences at the University of Alabama at Birmingham. His advisor is Dr. John Johnstone. His research interests include shape modeling and semantics, serious gaming, augmented and virtual environments and human computer interaction. His current work involves mesh decomposition and alignment.