University of California and Berkeley, United States
Floor Plan, Watertight Modeling, Range Data, LiDAR.
Augmented, Mixed and Virtual Environments
Computer Vision, Visualization and Computer Graphics
Geometry and Modeling
Modeling and Algorithms
Scene and Object Modeling
Automatic generation of building floor plans is useful in many emerging applications, including indoor navigation, augmented and virtual reality, as well as building energy simulation software. These applications require watertight models with limited complexity. In this paper, we present an approach that produces 2.5D extruded watertight models of building interiors from either 2D particle filter grid maps or full 3D point-clouds captured by mobile mapping systems. Our approach is to triangulate a 2D sampling of wall positions and separate these triangles into interior and exterior sets. We partition the interior volume of the building model by rooms, then simplify the model to reduce noise. Such labels are useful for building energy simulations involving thermal models, as well as for ensuring geometric accuracy of the resulting 3D model. We experimentally verify the performance of our proposed approach on a wide variety of buildings. Our approach is efficient enough to be us
ed in real-time in conjunction with Simultaneous Localization and Mapping (SLAM) applications.