High-Precision defect detection and geometric verification in pipe jacking projects using computer vision and point cloud data

Pipe jacking, a trenchless method for underground pipeline installation, minimizes surface disruption in urban environments but faces significant challenges in real-time quality control due to complex subsurface conditions. Existing approaches, primarily reliant on manual inspections or post-construction surveys, are labor-intensive, lack continuity, and often fail to detect defects or geometric deviations during construction. To address these issues, this study introduces a two-stage framework integrating computer vision and point cloud analysis for internal defect detection and geometric verification during construction.The methodology comprises two stages: Stage 1 involves a CCTV pipeline inspection robot equipped with a binocular-camera system and the YOLOv5 model for real-time detection and 3D localization of internal defects using stereo matching and SLAM. Stage 2 employs a GoSLAM handheld LiDAR scanner for continuous geometric verification through RANSAC fitting and ICP registration.Validated on a 2.58-km pipe jacking project in Harbin, China, the approach was capable of accurately identifying five common defect types—cracks, holes, leakage stains, sediment deposition, and spalling—achieving high overall precision and recall, with minor challenges in detecting less distinct defects like spalling. Geometric verification achieved a mean fitting error of 6.3 mm and an RMSE of 60.7 mm, with measured deviations—radial (16 mm), vertical (78 mm), horizontal (55 mm), and curvature (0.02°)—all well within permissible tolerances (159 mm, 112 mm, 0.5°). By enabling proactive monitoring and early defect correction, this dual-stage solution enhances the safety, efficiency, and reliability of pipe jacking construction, offering a scalable strategy for modern trenchless infrastructure projects.