Graphics and Media Lab

Introduction

This task is devoted to 3D-reconstruction of surfaces of revolution from user’s sketches on two images. Here is a novel approach which consists of three parts:

1. User-assisted selection of object’s contours on each image.
2. 3D-reconstruction of object’s axis of revolution using stereovision.
3. Modeling of cross-sections controlled by two images.

Algorithm description

First, user selects the left and the right contour of the object’s projection on each image. Then the axis of revolution is reconstructed according to the assumption that the axis of symmetry of the object’s projection on each image is a real projection of the object’s axis of revolution. We can get the upper and the lower points of the axis in space using stereovision. After the axis of the object is estimated, the surface of the object is reconstructed using baloon 3D deformable model [1].

The algorithm is remarkable for its stability and robustness. It means that a small change of input data doesn’t affect the result greatly. Some other algorithms of reconstruction of surfaces of revolution get model points isolately and this leads to scattering of points because of noisy data, truncation errors, etc. Our algorithm takes into account continious structure of the object and that's why if the input data is more or less correct it's impossible to get a wrong model.
The algorithm works well with all types of surfaces of revolution. Increasing the number of cross-sections of the growing cylinder we can obtain more accuracy because more detailed model can be reconstructed.

Results

Here are the results of the algorithm. From the left side and from the right side of each picture there are original images and in the middle there is a reconstructed and textured model.

Team

• Olga Senyukova, Lomonosov Moscow State University
• Dr. Vladimir Vezhnevets
• Anna Degtiareva

Publications

• Äåãòÿðåâà , Ñåíþêîâà . Ðåêîíñòðóêöèÿ òåë âðàùåíèÿ ïî äâóì èçîáðàæåíèÿì. GraphiCon , 2006 (download)