Extracting Bridge Components from a Laser Scanning Point Cloud

Abstract

A three-dimensional (3D) geometric model of a bridge plays an important role in inspection, assessment and management of the bridge. As most bridges were built after the second world war, 3D bridge models are rarely available. A recent development in laser scanning offers a cost-efficient method to capture dense, accurate 3D topographic data of the bridge. However, given the typical complexity of the bridge, a current workflow based commercial software to construct the bridge model still requires intensive labour work. This paper introduces a new approach to extract the point cloud of each surface of structural components of a slab/box beam bridge automatically in a sequential order from a superstructure to a substructure. The proposed method first employs a quadtree to decompose the point cloud of the bridge into two dimensional (2D) cells. Second, a kernel density estimation is used to separate a point cloud describing patches of surfaces within the cells. Subsequently, the cell- and voxel-based region growing are developed to segment patches within the cells/voxels for the superstructure and substructure, respectively. Moreover, knowledge of the bridge’s components (e.g. position, orientation, or shape) is introduced to allow the proposed method to identify criteria for filtering irrelevant objects, and to establish criteria for extracting the components. An experimental test shows the proposed method successfully extracts all surfaces of the bridge components.