3D Reconstruction with Prior Knowledge

Master's defence by Axel E. Jensen

It takes place: 29/8/2012 at 9:15 a.m. in SP25, parterre, Universitetsparken 1

Abstract:

Reconstructing 3D scene geometry from 2D images is usually done with no or few assumptions on the actual scene geometry. Usually because it is difficult to formulate such assumption or they don't exist. In some cases, a 3D model of the scene geometry does exist. This could be a 3D model of a building reconstructed from floor plans and height measurements. In this thesis it is explored how such prior knowledge can be used throughout the process of a 3D reconstruction.

Today, handheld devices, such as smart phones and tablets, are equipped with a range of sensors measuring their position, orientation and motion. These sensory measurements provide valuable information, which can be used in the 3D reconstruction process. A program for capturing sensory measurements and images, and subsequent processing, has been developed, and it is investigated how these sensory measurements can be incorporated into the 3D reconstruction solution.

Combining prior knowledge and sensory measurements facilitates 3D reconstruction of images taken from viewpoints far apart in location and orientation. Some tasks become easier to automate, such as choosing amongst different reconstruction techniques and avoiding degenerate configurations.

Supervisor: Søren Ingvor Olsen

Censor: Søren Overgaard