Figure 3.10 showed the initial image regions and Figure 4.8 shows the final surface hypotheses formed by the processes described in this chapter (numbers in the picture are the surface index).
There are several instances of successful surface reconstructions. Figure 4.9 shows reconstruction of the robot upper arm side panel. The surface merging operation has rejoined the two sections, and the boundary extension has largely restored the missing section in the lower left corner. Because the reconstruction is only suggestive, and because the leftmost boundary was not observed as being curved, the reconstruction is a little generous and slightly exceeds the actual surface. The right end of the panel could not be reconstructed because of insufficient evidence for the true boundary, although the labeling claimed it was obscured. Other valid reconstructions included the block lying on the table and the two halves at the back of the trash can (Figure 4.10). The latter is interesting because it was a curved surface, so matching of surface shape had to account for the change in orientation. One inappropriate merging occurred: the two top panels of the robot upper arm had their real join hidden behind the vertical gantry column. As they were continuous, their orientation differed only slightly, and met all the occlusion constraints, they were merged as well. The hypothesis construction process expects this type of error, so it did not prove catastrophic.
These examples show that the
obscured surface reconstructions are successful, and Figure
4.8 shows that most reasonable surfaces are made.