Automated quantification of crack length and width in asphalt pavements
Zhe Li,Tuo Zhang,Yi Miao,Jiupeng Zhang,Mehran Eskandari Torbaghan,Yinzhang He,Jiasheng Dai
DOI: https://doi.org/10.1111/mice.13344
IF: 10.066
2024-09-19
Computer-Aided Civil and Infrastructure Engineering
Abstract:Rapid, accurate, and fully automated estimation of both length and width of asphalt pavement cracks, essential for achieving a proactive asset management, presents a significant challenge, primarily due to limitations in the effectiveness of automatic image segmentation and the accuracy of crack width and length estimation algorithms. To address this challenge, this paper introduces the Branch Growing (BG) algorithm, specifically designed for crack length estimation in asphalt pavements, along with an optimized OrthoBoundary algorithm tailored for crack width estimation. Leveraging four widely adopted deep learning models for asphalt pavement crack segmentation, four distinct sets of image segmentation results have been produced. Subsequently, a comprehensive evaluation has been conducted to assess the effectiveness of both crack dimensions estimation algorithms. The findings demonstrate that the integration of the BG algorithm, the optimized OrthoBoundary algorithm, and the fully convolutional network with the HRNet backbone achieve a prediction accuracy of 80.21% for crack length estimation and 84.32% for average width estimation. Moreover, the image processing speed, at a resolution of 3024 × 3024, can be maintained at approximately 5 s, with average width estimation observed to be up to 9.1‐fold faster than the unoptimized OrthoBoundary algorithm. These results signify advancements in automated crack quantification methodologies, with implications for enhancing civil infrastructure maintenance practices.
transportation science & technology,construction & building technology,computer science, interdisciplinary applications,engineering, civil