Integrated Vision-Based System for Automated Defect Detection in Sewer Closed Circuit Television Inspection Videos

摘要

This paper discusses the development of a general framework and software system to support automated analysis of sewer inspection closed-circuit television (CCTV) videos. The proposed system aims primarily to support the off-site review and quality control process of the videos and to enable efficient reevaluation of archived CCTV videos to extract historical sewer condition data. Automated analysis of sewer CCTV videos poses several challenges including the nonuniformity of camera motion and illumination conditions inside the sewer. The paper presents a novel algorithm for optical flow-based camera motion tracking to automatically identify, locate, and extract a limited set of video segments, called regions of interest (ROI), that likely include defects, thus reducing the time and computational requirements needed for video processing. The proposed algorithm attempts to recover the operator actions during the inspection session, which would enable determining the location and relative severity of the ROI. To ensure proper segmentation and defect detection, frames within the ROI are classified on the basis of the camera orientation using a set of Haar-like features and a multiclass support vector machine. A segmentation algorithm based on gray-level intensity analysis is also presented. Algorithms for automated detection of debris and joint displacement defects are also discussed. The debris detection algorithm employs image segmentation and texture analysis techniques to locate and verify debris objects inside the water flow lines. The joint displacement algorithm performs gray-level intensity analysis to detect joints offset. The proposed system was successfully applied to analyze a set of CCTV videos obtained from the cities of Regina and Calgary in Canada. The results were validated against actual inspection reports prepared by CCTV operators, which demonstrated the viability and robustness of the proposed algorithms. (C) 2014 American Society of Civil Engineers.