3D model registration for image segmentation and texture mapping

Srimal Jayawardena, Di Yang and Jan Mechior (ANU, Ruhr-University)

COMPUTER VISION AND ROBOTICS SERIES

DATE: 2011-03-31
TIME: 15:00:00 - 16:00:00
LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU
CONTACT: JavaScript must be enabled to display this email address.

ABSTRACT:
Recovering the 3D pose from a generic object can be a daunting task. Especially when it is difficult to identify matching feature and texture between the 2D image and 3D model. Varying illumination conditions add further difficulty. We propose minimizing a loss that uses the image gradients in the given 2D image and image gradients of the projection of surface normal components of the 3D model to recover a full 3D pose. The projection of the 3D model at the recovered pose can be used to obtain a good initialization for segmenting the 2D image. Segmenting an object into components that share the similar features like color and texture is a non-trivial task. A good initialization obtained from the projected 3D model can help segmentation in such situations. We also use the recovered pose of the object to map the given 2D image as a texture on the 3D model. This can be used to apply transformations to the object in the 2D image, which are difficult with just 2D transformations. This also has useful applications in augmented reality (AR).

BIO:
Srimal Jayawardena - PhD Candidate at the Australian National University. Di Yang - PhD Candidate at the Australian National University. Jan Mechior - Intern at ANU, Masters student at Ruhr-University Bochum (Germany).