Attentive Point Aggregation Network for Semantic Segmentation of Building Pipeline Point Clouds

Abstract Accurate detection and modeling of pipeline systems from point cloud data are critical challenges in creating Building Information Models (BIM) of as-built environments. This study aims to develop an automated approach for semantic segmentation of building pipeline systems. To achieve this objective, we present two main contributions to the body of knowledge: a practical workflow for generating and annotating synthetic point cloud datasets from BIM models, which significantly reduces manual labeling effort, and the Attentive Point Aggregation Network (APANet), a novel deep learning architecture specifically designed for pipeline segmentation tasks. The effectiveness and components of the proposed network were analyzed through comprehensive ablation studies. Based on the synthetic dataset, experiments were conducted to evaluate the performance of APANet against baseline methods. Results demonstrated that APANet achieved performance with 92.62% overall accuracy and 85.10% mean Intersection over Union (IoU) when evaluated on real-world test datasets. Comparative evaluation against four baselines confirms that APANet achieves superior performance across all metrics on this domain-specific task. A supplementary evaluation on the Stanford Large-Scale Three-Dimensional (3D) Indoor Spaces (S3DIS) public benchmark further confirms the architectural generalizability of the proposed network. The ablation studies revealed that the feature aggregation module contributed most significantly to model performance, with a 12.86% improvement in mean IoU. The case studies further validated the effectiveness of the proposed approach in handling imbalanced point cloud distributions between cylindrical pipes and rectangular ducts. The cross-domain evaluation approach (training on synthetic data, testing on real data) and specialized network architecture presented in this study provide an efficient and task-specialized solution for point cloud recognition, segmentation, and modeling of different pipeline categories in building environments, which can significantly improve the efficiency and accuracy of as-built BIM modeling processes.