Automatic creation of architectural CAD models

This paper describes the automatic creation of architectural CAD files from a sequence of images acquired by a single camera mounted on a mobile robot. Segments with particular 3-D orientations are detected and tracked across several images using a matching algorithm. The 3-D position and orientation of the corresponding scene edges are reconstructed by a structure from motion algorithm. Next, an interpretation stage tries to identify planar patches in the 3-D data. These patches are likely to correspond to walls, ceilings and floors in the scene. The resulting scene description is converted to an industry-standard CAD file format (DXF) so that data can be manipulated or displayed using conventional CAD programs. The goal of the entire system is to reduce or eliminate human intervention in the generation of CAD descriptions of existing buildings. This paper focuses on the latter stages of the CAD model generation, including planar patch identification