Structure From Motion and Intrinsic Images

teaser

Figure 1: Left: A 3D point cloud reconstructed from images using Photogrammetry. Right: Result from the method [1] which separates reflectance from lighting.

This project involves combining 3D Reconstruction from images (Structure from Motion), and separation of illumination from colour (Intrinsic Images).

The student will investigate and implement, (some code already available), image processing techniques to separate lighting from colour from images. [1][2][3]

This techniques will be combined with 3D reconstruction from images [4][5][6]. A license of Agisoft[6] will be available, although source code from research solution is also available.[4][5]

Objectives
The main objective of the project is to achieve a working pipeline for producing 3D models with shadowless textures. The second objective is to research improvements about the data that can be captured, trying to include cases like specularity, small detail geometry, etc.

Basic Tasks

  • Get existing research code working [3][4]
  • Add functionality to export point clouds.[4]
  • Implement [2] (Re-implement matlab code) and combine with 4.
  • Combine [2] with Agisoft [6]
  • Research possible contributions.

References

  • [1] http://www-sop.inria.fr/reves/Basilic/2012/LBPDD12/CoherentIntrinsicImages_paper.pdf
  • [2] http://www-sop.inria.fr/reves/Basilic/2013/LBD13/LDB_TVCG12-paper.pdf
  • [3] http://giga.cps.unizar.es/~elenag/projects/EGSR2012_intrinsic/
  • [4] http://www.gris.informatik.tu-darmstadt.de/projects/multiview-environment/
  • [5] >http://research.microsoft.com/en-us/um/people/hoppe/proj/poissonrecon/
  • [6] http://www.agisoft.com/