Automatic CAD modeling of industrial pipes from range images

We present in this paper a method to obtain automatically CAD models of industrial pipes from range images. The models are based on two geometric primitives, cylinders and torii, which are enough to represent most parts of the pipes. The images are obtained with an accurate long-distance laser range sensor developed for reverse engineering in industrial structures. The key issue for automatic CAD modeling is the automatic segmentation of the data into subsets of points corresponding to the desired primitives. To do so, we use differential geometry to segment lines of centers of curvature into straight and curved parts, corresponding to the cylinder and torus parts of the original image. Differential geometry results being noisy and biased, we use an optimal approach for the computation of centers of curvature.