Dynamic indoor space reconstruction using temporal point clouds

Indoor positioning, route finding, and 3D modelling are essential techniques for indoor navigation. In indoor environments, users require the suitable route based on security, safety and efficiency, because an indoor navigation would be affected by various changing objects, such as pedestrian, escalators, and doors. Therefore, an indoor navigation requires a geometrical network model to represent changing space environments to be used for walkable path estimation. A geometrical network model is generally prepared using building blueprints or CAD data. However, there are technical issues such as high operation cost, because the geometrical network model is created manually. Moreover, a manual creation is hard to represent real-time environmental changes. Therefore, we aim to develop a methodology to provide the real-time geometrical space generation using temporal point clouds. The methodology consists of a point cloud interpolation, segmentation, clustering, and labeling to represent changing objects such as pedestrian and doors. We conduct experiments on dynamic indoor space reconstruction using multi-layered laser scanner to evaluate our methodology.