Abstract This paper presents a novel computer vision based system to support automated PACP (Pipeline Assessment Certification Program) coding for cracks. The proposed system comprises five major steps: 1) identifying forward facing view (FFV) with pipeline viewpoint detector, 2) obtaining stable edge information using structure edge detector, 3) acquiring crack segments and inner circle area of the pipeline with image binarization, 4) generating a 2D angular-diameter histogram for each frame, and 5) training a crack category classifier with support vector machine (SVM). The experimental results demonstrated that the proposed system can not only detect the cracks and categorize the crack type per PACP standard effectively but can also run about 10 frames per second (fps) in real world CCTV videos with 320 × 240 resolutions. In terms of accuracy in detecting cracks, the proposed method reaches about 91%, 88% and 90% recall for longitudinal cracks (CL), circumferential cracks (CC) and multiple cracks (CM), respectively. This paper contributes to the overall body of knowledge by providing an innovative framework that supports real-time crack identification and coding per PACP standards, which will lay a strong foundation for the development of a fully automated PACP sewer pipeline inspection system.

中文翻译：

---

闭路电视下水道检测视频小类裂缝分类

摘要 本文提出了一种新的基于计算机视觉的系统来支持自动化的 PACP（管道评估认证程序）编码裂缝。所提出的系统包括五个主要步骤：1）使用管道视点检测器识别前向视图（FFV），2）使用结构边缘检测器获取稳定的边缘信息，3）通过图像二值化获取管道的裂缝段和内圆区域，4 ) 为每帧生成二维角直径直方图，以及 5) 使用支持向量机 (SVM) 训练裂纹类别分类器。实验结果表明，所提出的系统不仅可以根据 PACP 标准有效地检测裂缝并对裂缝类型进行分类，而且还可以在具有 320 × 240 分辨率的真实世界闭路电视视频中运行约每秒 10 帧 (fps)。在检测裂缝的准确性方面，所提出的方法对纵向裂缝（CL）、周向裂缝（CC）和多裂缝（CM）的召回率分别达到了约 91%、88% 和 90%。本文通过提供一个支持实时裂缝识别和按照 PACP 标准编码的创新框架，为整个知识体系做出贡献，这将为全自动 PACP 下水道管道检测系统的开发奠定坚实的基础。