next up previous contents
Next: Conclusion Up: Some results Previous: Castle sequence   Contents

Fountain sequence

The Fountain sequence consists of 5 images of the back wall of the Upper Agora at the archaeological site of Sagalassos in Turkey, taken with a digital camera with 573x764 pixel resolution. It shows a concavity in which once a statue was situated.

Table 7.1: Statistics of the fountain sequence for visibility $V$, fill rate $F$ and depth error $E$.
N[view] $V[views]$ $F[\%]$ $E[\%]$
2 2 89.8728 0.294403
3 2.85478 96.7405 0.208367
5 4.23782 96.4774 0.121955


The performance characteristics are displayed in the table 7.1. The fill rate is high and the relative error is rather low because of a fairly wide baseline between views. This is reflected in the high geometric quality of depth the map and the reconstruction. Figure 7.19 shows from left to right images 1 and 3 of the sequence, the depth map as computed with the 2-view estimator, and the depth map when using all 5 images. The white (undefined) regions in the 2-view depth map are due to shadow occlusions which are almost completely removed in the 5-view depth map. This is reflected in the fill rate that increases from 89 to 96%. It should be noted that for this sequence a very large search range of 400 pixels was used, which is over 70% of the image width. Despite this large search range only few matching errors occurred.

Figure 7.19: First and last image of fountain sequence (left). Depth maps from the 2-view and the 5-view estimator (from left to right) showing the very dense depth maps (right).
\begin{figure}\centerline{
\psfig{figure=stereo/left.ps, width=6cm}
\psfig{figur...
...o/depth1.ps, width=6cm}
\psfig{figure=stereo/depth4.ps, width=6cm}
}\end{figure}


next up previous contents
Next: Conclusion Up: Some results Previous: Castle sequence   Contents
Marc Pollefeys 2002-11-22