An Efficient Three-Dimensional Point Cloud Segmentation Method for the Dimensional Quality Assessment of Precast Concrete Components Utilizing Multiview Information Fusion

Abstract

Significant discrepancies between the actual and designed dimensions of precast concrete (PC) components can result in construction delays and extensive rework, highlighting the imperative for dimensional quality assessment of these components. Point cloud technology can restore the three-dimensional (3D) information of objects, serving as an effective tool for dimensional quality assessment. The original point clouds often contain substantial nonessential data and outliers, necessitating the automatic extraction of relevant PC components from these complex point clouds. Existing point cloud segmentation methods suffer from insufficient accuracy and high cost. This study proposes a point cloud segmentation method employing multiview fusion for the dimensional quality assessment of PC components. First, an improved structure from motion (SfM) method is used to reconstruct the point cloud of PC components from multiview images. Second, an improved DeepLabv3 model is used to segment the multiview images and generate the corresponding masks. Third, a point cloud segmentation method using multiview fusion is proposed to extract the target PC component point cloud, facilitating the dimensional quality assessment. Compared to traditional point cloud segmentation methods (i.e., DBSCAN, K-means, mean shift, and region growing), the proposed method achieves the highest F-score exceeding 99.5%, and the relative errors between the calculated dimensions and the actual dimensions measured manually are below 1.3%. The results demonstrate the effectiveness of the proposed method in segmenting the point clouds of PC components and facilitating the dimensional quality assessment.