Enhancing Scan-to-BIM for reinforced concrete bridges using point cloud completion techniques

Abstract

Scan-to-BIM, the process of capturing the 3D point clouds and converting them into Building Information Models (BIM), is essential for modern bridge management systems. However, occlusion in point clouds poses significant challenges in designing reconstruction approaches and generating high-quality geometric BIM. This paper addresses this challenge by integrating a new point cloud completion module into the existing bridge Scan-to-BIM framework. The proposed module employs an improved point completion network (PCN) model to predict complete geometry from incomplete input, followed by using it to repair occlusions in incomplete point clouds. Its effectiveness was evaluated using both synthetic and real-world point cloud datasets. Experimental results demonstrated that (1) the proposed approach effectively resolves most occlusions in real-world datasets and (2) restores synthetic incomplete point clouds and enhances their geometric similarity to the ground-truth shapes, reducing Chamfer Distance (CD) by an average of 12.68 and increasing the F-score by 7.26 %.