3D Reconstruction from Images


The goal of multiview 3D reconstruction is to infer geometrical structure of a scene captured by a collection of images. The performed research (2006 - 2011) focuses on passive methods like multiview stereo and shape-from-silhouettes.


The conducted research exclusively studies variational formulations based on convex optimization where, starting from a generic initialization like a box, a surface evolves towards the final reconstruction by minimizing an appropriate cost functional. Unlike local optimization schemes, the developed convex models do not require an accurate initial surface estimate and confer a considerable degree of robustness.

3D reconstruction

Below, multiple reconstructions of ancient Greek statues are depicted, which were produced within a project with the Art Museum in Bonn, Germany. Generally, statues are well-behaved objects for multiview stereo approches as they exhibit sufficient material texture.

3D reconstruction3D reconstruction
3D reconstruction3D reconstruction
3D reconstruction3D reconstruction

Research efforts were focused on improving the accuracy of existing methods but also on increasing their robustness and extending the range of applicability. As a result, the developed approaches are able to operate in cases where traditional techniques fail (e.g. thin small-scale structures or lack of texture).

3D reconstruction3D reconstruction
3D reconstruction3D reconstruction