Acquisition of 3D models from pre-existing image sequences

Here the reconstruction of the ancient theater of Sagalassos is shown. Sagalassos is an archaeological site in Turkey. More results obtained at this site are presented in Sections 9.4 and 9.5. The reconstruction is based on a sequence filmed by a cameraman from the BRTN (Belgische Radio en Televisie van de Nederlandstalige gemeenschap) in 1990. The sequence was filmed to illustrate a TV program about Sagalassos. Because of the motion only fields -and not frames- could be used. The resolution of the images we could use was thus restricted to $768 \times 288$. The sequence consisted of about hundred images, every tenth image is shown in Figure 9.10. We recorded approximately 3 images per second.

Figure 9.10: This sequence was filmed from a helicopter in 1990 by a cameraman of the belgian television to illustrate a TV program on Sagalassos (an archaeological site in Turkey).

In Figure 9.11 the reconstruction of interest points and cameras is given. This shows that the approach can deal with long image sequences.

Figure 9.11: The reconstructed interest points and camera poses recovered from the TV sequence.

Dense depth maps were generated from this sequence and a dense textured 3D surface model was constructed from this. Some views of this model are given in Figure 9.12.

Figure 9.12: Some views of the reconstructed model of the ancient theater of Sagalassos.