Of particular interest is the problem of determining the
architectural structure of indoor environments (e.g., room walls,
floors and ceilings). Indoor reconstruction exhibits a number of
distinctive challenges that make it significantly harder to manage
than the more well-studied problem of building shape reconstruction from outdoor scans (see also Section 2). First of all, indoor
reconstruction methods must be significantly more tolerant to
missing data than their outdoor counterparts, since environments
such as offices and apartments exhibit extremely high levels of
clutter. This typically results in heavy occlusions of walls and other
structures of interest (see also Fig. 1). Secondly, windows and other
highly reflective surfaces are often present in such scenes. As a
result, the acquired model is heavily affected by large-scale
artifacts, measurement noise and missing data, due to the critical
interaction properties of the reflective elements with the measurement devices (see also Fig. 1). Finally, creating structured 3D
models of typical indoor environments, such as apartments and
office buildings, poses the challenge of recognizing their interior
structure in terms of a graph of connected rooms and corridors.