Refinement of coarse indoor models using position traces and a low-cost range camera

In our previous work, we described the automatic reconstruction of coarse models of building interiors using a single photograph of an evacuation plan and an available model of the building´s external shell. Additionally, we presented the extraction of the initial position and orientation for a foot-mounted MEMS IMU positioning system from the plan as well as the use of the coarse model for alignment and map-matching purposes. In this paper, we propose the geometric and semantic refinement of the coarse model employing an analysis of position traces delivered by such a system. While the addition of semantic attributes requires user interaction (like taking a photograph of a door plate to be analyzed by Optical Character Recognition), the update of the model´s geometry (i.e. reconstruction of doors) is delivered automatically by analyzing position traces and model conjointly. The model´s geometry is further refined using an automatic approach for the reconstruction of detailed CAD models of individual rooms, from the point clouds collected by the low-cost RGB-D sensor Microsoft Kinect. The user´s track derived from the MEMS IMU positioning approach will be used for the registration and fusion of the detailed models to the coarse model.