next up previous contents
Next: Volumetric integration Up: Texture enhancement Previous: Best view selection for   Contents

Example

We have recorded a short video sequence from a medusa head decorating an ancient fountain in Sagalassos (an ancient city in Turkey). The 20 second video sequence was recorded with a hand-held consumer video camera (Sony TRV-900). Each twentiest frame was used as a key-frame by our video to 3D processing pipeline. Three of these frames are seen on the top part of Figure 8.4. The compute structure and motion is also seen in this figure (middle-left). The camera viewpoints are represented by small pyramids. The depth map used to construct the 3D model is seen on the middle-right of the figure. Finally, the model -with and without texture- is seen at the bottom of the figure. From the shaded model one can see that most of the geometric detail is accurately recovered. By using the image itself as texture map a photorealistic model is obtained. Note from the rightmost view that the 3D model allows to render realistic views that are very different from the original views.
Figure 8.4: 3D model of a decorative Medusa head recorded at the ancient site of Sagalassos in Turkey. Top: 3 views of the original video, middle: reconstruction of 3D feature points with computed camera motion for the keyframes and one of the computed depth/range images, bottom: shaded and textured views of the recovered 3D model.
\begin{figure}\centerline{
\psfig{figure=mod/im.0300.ps,width=5cm}
\psfig{figure...
...od/medusa2c.ps,width=5cm}
\psfig{figure=mod/medusa3c.ps,width=5cm}
}\end{figure}


next up previous contents
Next: Volumetric integration Up: Texture enhancement Previous: Best view selection for   Contents
Marc Pollefeys 2002-11-22