Please use this identifier to cite or link to this item:
|標題:||Segmenting ideal morphologies of sewer pipe defects on CCTV images for automated diagnosis||作者:||Yang, M.D.
|關鍵字:||CCTV;Image processing;Morphologies of pipe defects;Diagnostic system;support vector machines;neural-network;component analysis;recognition;model;rehabilitation;segmentation||Project:||Expert Systems with Applications||期刊/報告no：:||Expert Systems with Applications, Volume 36, Issue 2, Page(s) 3562-3573.||摘要:||
Several literatures presented automated systems for detecting or classifying sewer pipe defects based on morphological features of pipe defects. In those automated systems, however, the morphologies of the darker center or some uncertain objects on CCTV images are also segmented and become noises while morphology-based pipe defect segmentation is implemented. In this paper, the morphology-based pipe defect segmentation is proposed and discussed to be an improved approach for automated diagnosis of pipe defects on CCTV images. The segmentation of pipe defect morphologies is first to implement an opening operation for gray-level CCTV images to distinguish pipe defects. Then, Otsu's technique is used to segment pipe defects by determining the optimal thresholds for gray-level CCTV images of opening operation. Based on the segmentation results of CCTV images, the ideal morphologies of four typical pipe defects are defined. If the segmented CCTV images match the definition of those ideal morphologies, the pipe defects on those CCTV images can be successfully identified by a radial basis network (RBN) based diagnostic system. As for the rest CCTV images failing to match the ideal morphologies, the failure causes was discussed so to suggest a regulation for imaging conditions, such as camera pose and light source, in order to obtain CCTV images for successful segmentation. (C) 2008 Elsevier Ltd. All rights reserved.
|Appears in Collections:||土木工程學系所|
Show full item record
TAIR Related Article
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.