|
|
Automatic reconstruction of video sequences from natural video sequences |
F. Galpin,
The goal of our work is the reconstruction of a natural video sequence
with a set of 3D models. The input video sequence is a set of images in the context of a static and rigid scene. This
issue is of interest in many applications, such as compression or navigation in
multimedia applications.
See also page 1.
This sequence is a part of a large indoor sequence. We can see the original one and the reconstructed one which uses several 3D models. Images of these sequences have been cropped in order to avoid flicking textures.
|
|
|---|---|
| Original video sequence | Reconstructed sequence |
This sequence is an outdoor sequence with sew geometric primitives. Optic flow algorithms are known to be better than primitive based algorithms on this kind of sequence. We can see the original one and the reconstructed one which uses several 3D models (this sequence have not been cropped, so we can see texture flicking). Moreover a texture morphing technique have been used in order to produce a smooth transition between 3D models. On bottom we can see a small part of the original sequence and a virtual reconstructed one : a rotation of 15deg over x axe have been applied on the camera.
![]() |
![]() |
![]() |
|---|---|---|
| Original video sequence | Reconstructed sequence | Zbuffer of a image in the sequence
(click to load a VRML model) |
![]() |
|
|---|---|
| Original video sequence | Reconstructed virtual sequence |