抄録
Mid-range and long-range laser scanners are useful to create point-based 3D models of manufacturing plants. Although point-based representation is promising for simulating renovation tasks of manufacturing plants, its discrete and unstructured nature prevents the user to efficiently edit point-clouds. Editing capability, such as deletion, displacement and addition of components and equipment, is very important for efficiently supporting renovation tasks. In this paper, we describe two editing technologies for point-clouds. First, we introduce a method for grouping points into meaningful sets. Our method consists of three stages; segmentation by continuous surfaces, segmentation by surface detection, and segmentation by object recognition. These techniques make it possible to efficiently edit point-cloud data. Then we introduce a new collision detection method for large-scale point-clouds. This technique is useful for simulating how to carry equipments out of or into a factory. For realizing efficient collision detection, we convert a point-cloud into a 2D depth map. Objects are projected into the depth map and are evaluated collisions by depth values on the depth map. We use multi-resolution depth maps in an out-of-core manner for handling large-scale point-clouds. In our experiments, our method enables to detect collisions in real time even when point-clouds are incomplete.
