Pipes are the basic building block in many industrial sites like electricity and chemical plants. Although pipes are merely cylindrical primitives which can be defined by axis and radius, they often consist of additional components like flanges, valves, elbows, tees, etc. 3D pipes are typically dense, consisting of a wide range of topologies and geometries, with large self-occlusions. Thus, reconstruction of a coherent 3D pipe models from large-scale point clouds is a challenging problem. In this work we take a prior-based reconstruction approach which reduces the complexity of the general pipe reconstruction problem into a combination of part detection and model fitting problems. We utilize convolutional network to learn point cloud features and classify points into various classes, then apply robust clustering and graph-based aggregation techniques to compute a coherent pipe model. Our method shows promising results on pipe models with varying complexity and density both in synthetic and real cases.
