Prometheus: From 2D to 3D. A reconstruction based on photographs.

On January 10th 2012 the statue of Prometheus, standing in front of the Aula, was stolen from the TU Delft. This statue was made specially for the TU Delft for its 100th anniversary in 1953. This thesis aims to reconstruct a digital 3d model of the stolen statue of Prometheus. We do this based on a set of photographs of the statue that were taken over the years. We use the SIFT feature detector to find keypoints in the photographs. By matching these keypoints between images we establish image correspondences. An incremental structure from motion approach is then used to create a projective reconstruction of the statue. The projective reconstruction is updated to a metric one by means of auto-calibration. We test our algorithm on three datasets. The first dataset contains several photographs of Prometheus. The second dataset contains the same photographs of Prometheus but with the background removed. The third dataset contains photographs from the preliminary study of the statue, these photographs are taken from around the statue, in the same lighting conditions and have guaranteed overlap. We will show that removing the background from the photographs before running our algorithm improves the reconstruction results. However, a full reconstruction of the statue seems impossible due to the lack of images and different conditions the photographs were taken in. If the conditions in which the photographs were taken in are similar and we have enough photographs, then our algorithm works well as we prove with the reconstruction results from the third dataset.

Subject

structure from motion
computer vision
3d reconstruction

To reference this document use:

http://resolver.tudelft.nl/uuid:7d2832e6-9921-4ad3-a925-0f17703a5894

Part of collection

Student theses

Document type

master thesis

Rights

Files

PDF

thesis-Prometheus_Irene_Renkens.pdf

15.19 MB

Close viewer