Learning deformable 3D models from multiple views

• Grant number: 09/00093-3
• Support Opportunities: Scholarships abroad - Research
• Start date: January 07, 2009
• End date: April 06, 2009
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Siome Klein Goldenstein
• Grantee: Siome Klein Goldenstein
• Host Investigator: Gabriel Taubin
• Host Institution: Instituto de Computação (IC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Institution abroad: Brown University, United States

Abstract

During this short visit, we will start a long project of collaboration where we will use computer vision, computer graphics, and artificial intelligence to learn tridimensional deformable models from multiple views. The first step toward this objective will use three types of information: silhouettes, depth-from-discontinuities, and dense stereo to reconstruct the volumetric hull of the model at every moment in time. Then, we will use develop a new set of algorihtms that will use the time-varying volumetric data as input to construct topologically consistent models over time, with properly parameterized dynamics and deformations. After that, the method can be then refined with the use of computational models of motion as constraints. Finally, the model needs a good representation, compression, transmission, and a good visualization for remote applications such as teleconference and sports and biomechanical data analysis. The goals of this first few months are clear: perform a literature review and write a survey, prepare two grant proposals (NSF and HP), and build a 3D capture system based on dense stereo using the prototype developed at Brown. (AU)