The integrity of drinking water pipelines is essential for ensuring public health and conserving increasingly scarce water resources. Internal inspection of pipelines enables non-destructive integrity assessment and informed maintenance decision-making for water infrastructure operators. Laser Projection Profilometry is a cost-effective, high-resolution approach for acquiring geometric information of inner pipeline surfaces.
This work proposes a comprehensive data processing pipeline for Laser Projection Profilometry, which enables 3D reconstruction and defect-oriented segmentation of pipeline interiors using a single-camera sensor configuration. The pipeline consists of camera calibration, image preprocessing, laser line extraction, 3D reconstruction in image coordinates, point cloud preprocessing, and unsupervised clustering-based segmentation. Individual image-sourced point clouds are spatially registered in 3D using estimated laser profile circle centers and stacked in the longitudinal direction. Quantitative evaluation is conducted using quasi-synthetic reference 3D point clouds generated from real measurement inputs, allowing controlled variation of pipeline geometry and defect characteristics while maintaining realistic sensor conditions.
Experimental results obtained from multiple real-world datasets demonstrate that the proposed pipeline achieves mean point-wise registration errors of 1.08–2.08 px, corresponding to 0.43–1.79 mm when converted using dataset-specific pixel-to-millimeter calibrations. The corresponding HD95 values range from 2.50–3.91 px and from 0.99–3.15 mm across datasets. Variations in both error metrics are primarily attributable to differences in pipeline diameter, surface material, and laser emitter intensity across the datasets. The method reliably reconstructs and localizes both positive (material buildup) and negative radius (material loss) defects across varying pipeline geometries under controlled inspection conditions. Density-based clustering (DBSCAN) proved robust and computationally efficient for defect segmentation, exhibiting real-time performance, whereas a hierarchical pruning approach (Torque Clustering) required careful parameterization and incurred a significantly higher computational cost. Identified limitations include material-dependent reflectance effects, geometric reconstruction of negative radius defects, sensitivity to robot orientation, and scalability constraints in hierarchical clustering. Overall, this work demonstrates the suitability of a Laser Projection Profilometry–based data processing pipeline for internal drinking water pipeline inspection and provides a quantitative foundation for future research toward more robust, scalable, and deployable inspection systems.