Pavement Crack Classification Based on Tensor Factorization
Date
2012
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Cracks in pavements allow water to pervade the various layers reducing drastically its ability to perform the primary function of carrying traffic loadings. The various types of cracks that occur on road pavements raise concerns for engineers and infrastructure managers. Identifying the type of crack accurately and efficiently is essential in road maintenance as this will lead to the prescription of cost-effective maintenance and treatment procedures. In the past, various image processing techniques have been applied in the detection and classification of pavement cracks most of which employ machine learning methods. This thesis outlines the importance of tensor analysis and decomposition as an alternative means of pavement crack classification. Tensors are multidimensional arrays and are generalizations of scalars, vectors and matrices. Two main types of cracks; longitudinal and transverse cracking are considered in the study. Due to the nature of tensors, the training set of images used is analyzed in a 3-dimensional space which captures variation across all images and ensures a more robust tensor algorithm for accurate crack classification. The levels of accuracy obtained after using the algorithm implies that crack classification based on tensor decomposition is one that can be successfully employed by state agencies nationwide who use digital image processing systems as part of their pavement management programs.