There are different types of occlusions which may arise when a scene is scanned with a laser beam or structured light sensor. Lots of these occlusions are not resolvable without domain specific knowledge, or models, helping to derive structural interpretation of the images. For this work, we only consider indoor scenes containing mostly man-made objects, for which both general architectural and scene specific knowledge might be available.
Previous research on occlusion reconstruction focused on the reconstruction of a single large area occluded by one object. In that context two cases were considered: Occlusions Preserving Surfaces, Occlusion Breaking Surfaces.
The detection and the reconstruction of the occlusions is based on searching for regions entirely contained within the boundaries of another region, so that they can then be extended across the occluding area [5].
The detection and the reconstruction is based on identifying compatible regions which can then be merged in order to reconstruct the missing occluded part [1, ].
In both the above cases the occluding object does not obscure the boundaries of the occluded one. The work of [9] explored the case when the occluding part partially obscures boundaries of an occluded object. This case is called Occlusion Breaking Boundaries. The corner area of a cupboard occluded by an open door, as in figure 1, is an example of this case. Figure 2.c shows an example of these occlusions.
The detection and reconstruction is based on establishing a foreground - background relation between regions during the analysis of region boundaries. In particular, the background region is occluded by the foreground region and reconstruction starts from its boundary.
In order to reconstruct occlusions we need to exploit available information that constrains a scene. In particular:
1) good surface continuation. That is, the occluded surface keeps the same shape of its visible part [8, 12].
2) good boundary continuation. That is, the occluded boundary keeps the same slope of its visible part.
3) architectural constraints. That is, the occluded surface is bounded by an architectural constraint. This can be a wall, a floor, a door, a window etc. [3, 7].
The reader should note that in order to solve cases of Occlusion Preserving Surfaces and Occlusion Breaking Surfaces, the surface good continuation constraint suffices. In the proposed case of Occlusion Breaking Boundaries it is also needed to apply the constraint of good boundary continuation and the architectural constraints.