The detected surfaces are considered in pairs and their boundaries compared. If depth values associated with adjacent surface boundaries are similar, the surfaces are contiguous and so not occluded in the area surrounding those boundaries. The boundaries between the wall and cupboard shown in figure 6 are examples of boundaries between contiguous surfaces. We call them true boundaries.
If instead, depth values associated with adjacent surface boundaries are different, it means that the surface closer to the sensor is in foreground and the other one is in background. In this case the boundary of the background surface is a false boundary. In figure 6 the boundaries between door and cupboard are examples of false boundaries. We have to establish a foreground-background relation for each pair of adjacent regions.
Figure 6: Endpoints and boundaries between adjacent surfaces.
Once false and true boundaries have been identified it is possible to detect the points where the background surface boundary is occluded. We call them endpoints. The endpoints are the ends of the background surface true boundaries and they lie just next to the false boundary endpoints. Figure 6 shows the endpoints for the example case. It is important to detect endpoints because they represent the starting points for the reconstruction.