Toward Automated Modeling of Floor Plans

This paper describes an automated method for creating accurate 2D floor plan models of building interiors. Our approach takes as input a 3D point cloud, which is obtained from laser scanners positioned at multiple locations in the facility. We use a histogram of height data to detect floor and ceiling data. We then project the remaining points onto a 2D ground plane and create a histogram of point density, from which line segments corresponding to walls are extracted using a Hough transform. Openings due to doors and windows arise implicitly from the modeling process. Our algorithm operates in unmodified, cluttered environments. We also propose a set of evaluation measures for objectively measuring floor plan modeling algorithm performance. Finally, we propose strategies for mitigating the effect of clutter through strategic selection of cross-sections containing minimal clutter. Our experiments on data from a 40-room school building achieve promising