3D classification and modeling of point clouds is an important but time-consuming process, with many applications in petroleum engineering and other applications, inspiring extensive research in automatic methods. Prior efforts focus on primitive geometry, street structures or indoor objects, but industrial data has rarely been pursued. Industrial sites are often captured by laser scanners, producing 3D point clouds of visible surfaces. Creating a higher level model from such point clouds now requires extensive manual labeling and fitting of surfaces and their connections. Computer-aided editing systems and automatic methods exist, but with limited capabilities. Existing methods for industrial point cloud modeling are not able to effectively process an industrial scene containing primitive shapes, general objects and instrumentations, as well as the connectivity between them.

Our work presents a method which automatically generates a 3D classified model from 3D point cloud data. An industrial point cloud is treated as a collection of pipes, planes and general objects, and we thus divide the modeling problem into three sub-problems: pipe modeling, plane classification, and object recognition. Two primitive extraction processes detect cylinder or planar geometry in the scene, while general objects modeling employs modeling-by-recognition, which detects clusters of 3D points in a scan that match 3D models of objects stored in a prebuilt object library. The results of all three components, including extracted primitives and recognized objects, are integrated to obtain the complete 3D model.

Experiments with several point cloud datasets demonstrate that the presented method automatically produces classified models of large and complex industrial scenes, with a quality that outperforms leading commercial modeling software and is comparable to professional hand-made models. Pipes, planes and different types of objects are individually detected and classified. The result display is freely switchable between mesh model for efficiency and point cloud for accuracy, not achievable in simple surface fitting methods.

You can access this article if you purchase or spend a download.