Monocular 3D Reconstruction of Poorly-textured Deformable Surfaces

Mathieu Salzmann (NICTA/ANU)

COMPUTER VISION AND ROBOTICS SERIES

DATE: 2012-07-12
TIME: 16:00:00 - 17:00:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.

ABSTRACT:
Abstract: Most recent approaches to monocular non-rigid 3D shape recovery rely on exploiting point correspondences and work best when the whole surface is well-textured.In this talk, I will present two approaches to performing 3D reconstruction when these assumptions are violated. First, I will introduce a method that exploits shading to reconstruct partially textured surfaces under complex lighting conditions. The key ingredients of this method are learned mappings from intensity patterns to local surface shapes and a principled approach to piecing together the resulting local shape estimates. Second, I will introduce a method inspired by discrete optimization techniques. I will show that non-rigid 3D reconstruction can be formulated as inference in a graphical model, which allows to exploit complex image likelihoods even in the +absence of a good initial shape estimate.

BIO: