Point Cloud–Based Defect Inspection with Multisensor Fusion and Deep Learning for Advancing Building Construction Quality

Abstract

Building defects are critical because they compromise safety, raise costs, and cause delays in construction projects, ultimately affecting the quality and structural integrity of buildings. However, traditional methods for identifying defects, which primarily rely on manual inspections, tend to be time-consuming and error-prone. To address this issue, we propose a novel defect inspection system for building construction that leverages the fusion of multiple red green blue-depth (RGB-D) sensors and deep learning to enhance accuracy and on-site applicability. The method consists of three steps: (1) point cloud data acquisition via multisensor fusion; (2) deep learning-based point cloud registration for generating a point cloud map; and (3) defect inspection from point cloud data and defect visualization on the point cloud map. This approach facilitates detailed analysis of structural defects, including framework distortions and sagging of ceilings and floors. Experimental results validated the system’s ability to inspect structural defects in buildings effectively, offering a promising tool for construction managers to identify structural issues in advance and implement corrective measures to enhance the overall quality of the building.