Fast, Automated, 3D Modeling of Building Interiors

Abstract : Fast, automated, photo-realistic 3D modeling of indoor environments is useful in military reconnaissance applications and civilian emergency response applications. Over the past five years, we have developed a human operated ambulatory backpack system equipped with 2D laser range finders, IMU and cameras in order to capture geometry and texture of the building interiors. The main advantage of our approach over a wheeled system is its capability to model uneven surfaces such as staircases; however our system requires 6 DoF pose recovery, rather than 3 for wheeled systems. As such, we have developed a number of sensor fusion and localization algorithms to recover full 6D pose and characterized their performance and accuracy using a highly accurate navigation grade IMU HG9900. An important component of our localization algorithms is loop closure detection and enforcement algorithms for which we use camera imagery as input. Localization of the data acquisition system enables generation of 3D point clouds, which are then input to our watertight surface reconstruction algorithms such as voxel carving, or 2D floor plan recovery algorithms. The models are then texture mapped with camera imagery using our algorithms which are designed to minimize seaming artifacts between images.