A crack detection and diagnosis methodology for automated pavement condition evaluation

Pavement condition evaluation;Crack detection;Crack classification

Jiang, Chenglong ; Tsai, Yichang Civil and Environmental Engineering Amekudzi-Kennedy, Adjo A. Lai, James Meyer, Michael D. Wang, Zhaohua Yezzi, Anthony J. ; Tsai, Yichang

University:Georgia Institute of Technology

Department:Civil and Environmental Engineering

关键词: Pavement condition evaluation;    Crack detection;    Crack classification;   

Others  :  https://smartech.gatech.edu/bitstream/1853/55524/1/JIANG-DISSERTATION-2015.pdf

美国|英语

来源: SMARTech Repository

【 摘 要 】

Pavement cracks are one of the most common pavement surface distresses. Traditionally, pavement crack evaluation is conducted manually through a human field survey, which is subjective, labor-intensive, time-consuming, and dangerous to field engineers in the hazardous roadway environment. Methods have been developed to automate this process, including both crack detection and classification. However, none of these methods have been fully automated and implemented. For crack detection, the active contour based minimal path procedure has demonstrated better performance than other existing algorithms; however, it mathematically requires one or more prior input points for each crack curve and cannot be fully automated, which introduces tedious and subjective manual work in practical use, especially for high-density cracks. In this study, a crack detection algorithm to detect accurate and continuous crack curves in a fully automatic manner is proposed. It consists of five major steps: 1) image enhancement through non-crack feature removal and profile rectification, 2) crack potential map generation through adaptive thresholding and tensor voting, 3) key point identification through iterative morphological thinning and skeleton analysis, 4) crack curve detection between neighboring key points using two-point minimal path procedure, and 5) crack curve selection through statistical analysis to remove false positive artifacts. The proposed algorithm has demonstrated its robust performance with a diverse data set, including the different pavement surface types (asphalt dense graded surface, asphalt open graded friction course surface, and concrete surface), and the different crack patterns (longitudinal, transverse, the combination of both, and alligator). Also, the proposed algorithm has outperformed the dynamic optimization algorithm, which has had the best accuracy as an automatic algorithm in pertinent literature, and the commonly used commercial crack detection software. For crack classification, although methods have been developed to classify the primitive crack types (e.g. longitudinal and transverse cracks), it remains a challenge to classify the complicated, real-world crack types and severity levels specified by transportation agencies. In this study, a generalized crack diagnosis framework is proposed to transform detected crack maps into meaningful decision-support information valuable to engineers. The proposed framework features a multi-scale crack representation based on a crack fundamental element model, which extracts rich properties from the detected crack maps, and a supervised classifier calibrated with real pavement data, which interprets the extracted crack properties into crack types and severity levels can be flexibly applied to different distress protocols. The proposed framework has been implemented with real-world distress protocols, including AASHTO standard PP 67 and GDOT's PACES protocol. Experienced pavement engineers have compared the results against manual surveys. The proposed framework has demonstrated over 90 percent crack classification and quantification accuracy. Overall, the proposed crack detection and diagnosis methodology has made significant progress in developing an automated pavement crack evaluation method.