Enhancing pipeline defect detection under non-shutdown conditions using cylindrical 2D transformation of monocular endoscopic single-frame images

초록

Water pipeline systems are vital infrastructure supplying essential resources for daily life, requiring continuous maintenance to ensure water quality. However, assessing these underground supply pipelines remains highly challenging. While sensor-based technologies like CCTVs, laser scanning, and sonar are available, CCTV-based visual inspection is most widely used for detecting a wide range of defects. To overcome the limitations of manual inspection, such as large footage volumes and subjective evaluation, automated computer vision and AI-driven defect detection technologies have emerged. Advanced robotic systems integrating various sensors can reconstruct pipeline structures, but their application is heavily restricted in small-diameter (100-150 mm) pipelines that remain in constant operation. To address these limitations, this study proposes a novel technique for estimating camera pose and mapping pipeline interiors using vanishing points extracted from single frames of a monocular camera. Unlike conventional methods, the proposed mapping technique reconstructs internal structures with actual unit resolution. This enables quantitative evaluation and effectively supports data-driven decision-making for pipeline maintenance.

키워드