Important: The informations on this page will probably be subject of change. This will be communicated during labs sessions.
The goal of the work during the whole term is to reconstruct a 3D object (scene) from its images. In order to make the task manageable, we have chosen such a scene, that is relatively uncomplicated considering 3D computer vision methods: a decorative portal. The simplicity of such a scene lies in the fact, that the scene is almost planar and small number of views is enough for reconstruction.
We have prepared data captured at several places, see Data. The capturing scheme consists of three levels of height, four pictures in each. The capturing scheme is in figure 1. All images are captured with the same zoom setting, i.e. the same internal camera calibration matrix K. The matrix is provided in each data set.
Since the 3D reconstruction algorithms use the (linear) perspective camera model, the radial distortion of input images has been removed.
The sparse correspondences for provided scenes has been computed and they are available. Note, that the correspondences are tentative, so they contain also mismatches.
The correspondences are stored in several files:
u_<id>.txt
, one row per coordinate pair x y
, where <id>
is the image identifier (two digit number)
m_<i1>_<i2>.txt
, one row per pair of indices, where <i1>
and <i2>
are identifiers of two images, i1
< i2
Example:
The file m_05_06.txt
begins with,
4 7 5 18285 11 27631 ...
the file u_05.txt
begins with
5.3 1613.4 7.0 364.8 9.5 1522.3 9.9 585.1 10.9 571.7 <--- 11.2 578.6 11.3 666.1 ...
and the file u_06.txt
begins with
6.3 1749.0 8.4 1753.3 8.9 497.9 10.4 540.9 11.0 683.2 11.0 687.8 11.1 589.8 11.3 583.4 <--- 12.1 1212.6 12.2 949.3 ...
The first row in m_05_06.txt
means, that the point 4 from the image 05
and the point 7 from the image 06
corresponds. I.e., the point with coordinates x=10.9, y=571.7
in the image 05
corresponds to the point x=11.3, y=583.4
in the image 06
.