Automatic Detection of Cylindrical Objects in Built Facilities

Abstract

Three-dimensional (3D) facility models are in increasing demand for design, maintenance, operations, and construction project management. For industrial and research facilities, a key focus is piping, which may comprise 50% of the value of the facility. In this paper, a practical and cost-effective approach based on the Hough transform and judicious use of domain constraints is presented to automatically find, recognize, and reconstruct 3D pipes within laser-scan-acquired point clouds. The core algorithm utilizes the Hough transform’s efficacy for detecting parametric shapes in noisy data by applying it to projections of orthogonal slices to grow cylindrical pipe shapes within a 3D point-cloud. This supports faster and less-expensive built-facility modeling. It is validated using laser-scanner data from construction of the Engineering-VI building on the University of Waterloo campus. The system works on a typical laptop. Recognition results are within a few millimeters to centimeters accuracy in accordance with the chosen tessellation of the Hough space. Broad applications to pipe-network modeling are possible.