| Cracking is one of the major and common highway pavement distresses; itaccelerates pavement structural deterioration, and negatively impactsthe overall highway performance and lifespan. Considering the growing demand for pavement maintenance, automatic pavement cracking detection is of great practical value for pavement maintenance and management, pavement performance evaluation and prediction, and materials and structure design.Pavement cracking includes features of location, type, intensity, size and so on. Since pavement condition is of complexity and variety, pavement crack detection is mainly confronted with challenges as follows:(1)pavement cracking can hardly be detected rapidly, precisely, completely and robustly; (2)a systematic evaluation mechanism of crack recognition algorithms is in need; (3)there is a lack of any algorithm effectively classifying pavement crack types and intensities. In this dissertation, research progresses are reviewed about automatic pavement cracking detection. Research on automatic pavement cracking recognition, pavement cracking recognition algorithmevaluationand automatic pavement crack classification is presented in depth based on lmm/pixel high resolution 3D laser images of pavements, aiming at the goal of obtaining pavement crack information rapidly, precisely, timely and effectively. The main work and achievements of this dissertation include:1. Considering existing image denoising and enhancing techniques are mainly focusing on pavement grayscale images, pre-processing methods suitable for high resolution 3D laser pavement images are proposed to eliminate preliminary noises and improve image quality, whichincludes median raphe removal algorithm and dimension reduction processing. They lay the foundation for pavement crack recognition and classification. In median raphe removal processing, median raphe of a pavement image is removed according to height difference betweenthe left and right half images. In dimension reduction processing, a source image is divided into blocks of 8 pixels × 8 pixels, thus a lower dimension image is formed, on which subsequent pavement cracking recognition algorithms are operated directly.Test results show that median raphe removal and dimension reduction processing can reduce impacts of median raphes, noises and textures, decrease computation cost dramatically and be beneficial for automatic pavement cracking recognitionin later stages for improvements in processing speed and accuracy.2. Current automatic pavement crack recognition algorithms can hardly identify complete crack images with fast processing speed and high accuracy, and they are mostly designed based on pavement grayscale images, thus a parallel crack detection algorithm is proposed based on crack seeds recognition regarding high resolution 3D laser pavement images, for the purpose of improving processing speed and recognition accuracy. The algorithm has 10 parallel sub-workflows, and each of them includes three key steps, crack seeds recognition, crack connection, crack seeds fusion and denoising processing. Finally,10 potential crack images from the parallel sub-workflows are combined and denoised, from which crack image is obtained. Test results show that the parallel framework substantially improves processing speed and crack fusing, as a result, the proposed algorithm performs well in terms of speed, precision and recall, and outperforms OTSU segmentation and Canny edge detection.3. Existing pavement cracking recognition algorithms are mostly on the basis of 2D image processing. In order to recognize pavement cracking more robustly and effectively using 3-dimensional image processing techniques, an automatic 3D virtual pavement crack detection algorithm is proposedconsidering 3-dimensional characteristics of pavement. First, four lighting intensity images are generated by lighting model processing on a source 3D laser pavement image at four angles. Second, cracking of the source 3D laser pavement image is detected using the proposed parallel crack detection algorithm based on crack seeds recognition. Then, shadow and reflection regions are detected as possible cracks from the four lighting intensity images. At last, crack detection results from the previous two steps are combined and denoised to form final crack image. Test results show that the proposed algorithm has a high level of crack fusion and good sensitivity, and it performs well at precision, recall and robustness.4. Aiming at performance evaluation, a systematic mechanism is proposed for objectively and reasonably evaluating, comparing and analyzing performanceof pavement crack detection algorithms, for (1)we do not have benchmark images to test and compare algorithm performance, (2)existing evaluation indicators are not reasonable enough, (3)we are short of theoretical basis for algorithm parameter determination and (4)existing evaluating methods have a limited scope of application. Correspondingly, the systematic mechanism involves four elements:(1)establish a benchmark library of test images of pavement crack detection, (2)modify performance evaluation indicators, (3)propose a parameter determination approach and (4)improve performance evaluation methods. Performance evaluation results based on the mechanism can be a guidance to improve crack detection algorithms.5. Since there’s no effective and automatic approach for identifying pavement crack types and intensitiesrecently, an automatic classification and intensity recognition algorithm of pavement cracking is proposed using image processing based on crack image vectorization and geometric feature extraction. First, cracking areas are automatically separated on the basis of contour vectorization in case that a pavement image has several cracking targets. Second, calculation formulas for 9 crack features are presented, among which cavity, length-width ratio and orientation significance are chosen as crack classification features. Then, a thesholding classification algorithm for distinguishing linear and netted cracks is designed. Precision and recall of the classification algorithm are assessed and tested. At last, intensities of cracks are recognized according to crack width and lumpiness features. Test results show that the proposed classification algorithm can identify pavement crack types and intensities of test imagesin real time without human intervention at high levels of precision and bias.Broadly speaking, contents above form an integrated system to obtain pavement crack information, whichprovides a reference for decision making of pavement maintenance and management, pavement performance evaluation and prediction, and materials and structure design. Part 1 pre-processes pavement 3D laser images and lays a foundation for pavement crack recognition and classification. Part 2 and 3 conduct deep research on pavement crack recognition (crack positioning). Part 2 recognizes cracking using 2D image processing techniques, while part 3 employs 3-dimensional characteristics of pavement to detect cracks, which is beyond plane image processing. Part 4 proposes a systematic mechanism for evaluating and analyzing performanceof pavement crack detection algorithms, and it points out the direction of the algorithm improvement. Part 5 extracts pavement crack types, intensities, sizes, etc., based on crack recognition results. |
PDF Full Text Request