Semi‐supervised pipe video temporal defect interval localization

Abstract

In sewer pipe closed‐circuit television inspection, accurate temporal defect localization is essential for effective pipe assessment. Industry standards typically do not require time interval annotations, which are more informative but lead to additional costs for fully supervised methods. Additionally, differences in scene types and camera motion patterns between pipe inspections and temporal action localization (TAL) hinder the effective transfer of point‐supervised TAL methods. Therefore, this study presents a semi‐supervised multi‐prototype‐based method incorporating visual odometry for enhanced attention guidance (PipeSPO). The semi‐supervised multi‐prototype‐based method effectively leverages both unlabeled data and time‐point annotations, which enhances performance and reduces annotation costs. Meanwhile, visual odometry features exploit the camera's unique motion patterns in pipe videos, offering additional insights to inform the model. Experiments on real‐world datasets demonstrate that PipeSPO achieves 41.89% AP across intersection over union thresholds of 0.1–0.7, improving by 8.14% over current state‐of‐the‐art methods.