Automated recognition of 3D pipelines from point clouds

Abstract

This study proposes a method for detecting and reconstructing pipelines from a 3D point cloud. First, the method extracts points on cylindrical objects using various properties computed with the principal curvatures. Next, the possible radii of the cylinders in the point cloud are estimated using a histogram constructed with the radii of the curvature at each point. Once the candidate radii are obtained, spheres are estimated using a RANdom SAmple Consensus-based algorithm, whose centroids are processed to find the orientation and centerline of each cylinder. The nearest cylindrical components which are detected are then analyzed to establish connectivity to determine how they are arranged in space. Depending on the type of connectivity, elbows and/or T-junctions are used to connect the cylindrical elements to form pipelines. The proposed method was tested with synthetic and scanned point clouds and demonstrated better performance than that of the existing methods.