Closed-circuit television (CCTV) is a common method for drainage pipe inspection and currently, the interpretation of CCTV images is mostly conducted manually. In this study, an integrated algorithm, namely DeepLab V3 Plus-Crack Length Quantification (DL-CLQ), is proposed to detect and quantify pipe cracks, which combines a semantic segmentation model and the newly developed Crack Length Quantification algorithm. DeepLab V3 Plus models with MobileNet V2 or Xception-65 backbones are trained for semantic segmentation and their performance is compared. Then, the Crack Length Quantification algorithm, integrated with Hough circle detection method, is utilized to quantify the length of cracks. A correct factor is introduced considering the effects of camera angles and low-resolution images. The proposed algorithm is verified by artificially created pipe cracks. The lengths of these cracks are set to be 30 mm, 60 mm and 100 mm. 4 different images were selected for each type of crack, and the averaged value calculated from the 4 images by the proposed algorithm was used for validation. In the artificial scenarios, DL-CLQ shows a mIoU higher than based-line models in segmentation, and lower in MSE of crack quantification, which indicates the accuracy of crack quantification depends greatly on the segmentation accuracy. This study provides an innovative method for automatic drainage pipe defect detection and quantification, and can also contribute to the further development of smart management for urban drainage networks.
