Pavement crack detection on geodesic shadow removal with local oriented filter on LOF and improved Level set

• Published in: Construction & building materials 2020-03, Vol.237, p.117750, Article 117750
• Main Authors: Wang, Weixing, Li, Hongxia, Wang, Kevin, He, Changtao, Bai, Minghui
• Format: Article
• Language: English
• Subjects: Binary images; Clustering algorithms; Crack detection; Edge detection; Geodesic shadow removal; Geodesy; Image enhancement; Image segmentation; Image segmentation algorithm; Level set; Local oriented filter; Local outlier factors; Minimum spanning trees; Oriented filter; Pavement crack detection; Pavements; Road crack; Road cracks; Shadow removal; Trees (mathematics)
• ISSN: 0950-0618; 1879-0526; 1879-0526
• DOI: 10.1016/j.conbuildmat.2019.117750

•Geodesic shadow removal algorithm is used to remove shadow and noise.•Local oriented filter algorithm on LOF is applied to enhance the crack information.•Algorithm on improved level set with crack indication is to extract the crack edges.•Post functions for noise removal, and short and long gap fillings in a binary image. In order to detect cracks on pavement accurately and quickly in an image, a new algorithm is proposed to combines geodesic shadow removal (GSR), local outlier factor (LOF) and improved Level set algorithms. It includes four steps: (1) GRS algorithm and Gaussian filter are used to remove shadow and noise; (2) the local oriented filter algorithm based on LOF is applied to enhance the crack information and suppress image noise; (3) the algorithm based on improved level set combined with crack indication function is utilized to extract the crack edges, and it has the self adaptability and the higher segmentation accuracy than an ordinary level set algorithm; and (4) in the binary image, a number of post processing functions are made for noise removal, and short and long gap fillings. In algorithm comparison, we tested different kinds of pavement crack images, and for the images, we compared several traditional image segmentation algorithms, such as Canny operator, FCM, Adaptive threshold, Minimum Spanning Tree (MST), Clustering analysis and CV model etc. The experimental results show that the proposed algorithm can improve the image segmentation accuracy and is effective in extracting pavement cracks.