Keywords: pattern recognition, classification, crack, integrity assessment, bridge management, self-organizing map, image processing.

At present day, public infrastructures such as bridges and roads can be seen all around. The continuous construction of such infrastructures is posing an important problem of how to maintain and efficiently manage an existing structure. It does not make much sense economically to demolish and rebuild all such structures for the mere reason that their damage is progressing. Therefore, it is necessary to develop a system that enables us to make correct judgements of which structures should be rebuilt and which ones should be repaired. While such judgements would not be difficult if structures present visually observable, apparent damage, it would be quite difficult if they contained damage inside, showing no trace of it on the outside. To make proper judgements on complex damage levels, it is necessary to bring visually observable damage into relation with various test-results for assessing damage levels as well as experts' mutual knowledge. However, because available data for the assessment of integrity is not sufficient both in quality and in quantity, we have to utilize the insufficient data from various aspects that we have and estimate the degree and the advancement of damage based on intuition and experience. Thus, integrity assessment has been dependant upon intuition and the judgements of experts. On the other hand, with increasing need of maintenance and repairs of these structures, the shortage of experts is only growing.

In this paper, the system of extracting characteristics of cracks showing up on concrete slabs through digital images is described and classification based on damage levels is attempted by using these results. First, the linear pattern of cracks is extracted from the digital images of the concrete slabs through image processing techniques. Next, the characteristics such as the projection histograms, that is often applied in the field of optical character recognition, and the feature points in the border expression are extracted. Finally, the digital images of cracks are classified into different damage levels based on the extracted characteristics through the LVQ (Learning Vector Quantization) system.

Cracks in a digital image of concrete slab were detected in accordance with the following procedures. The digital image underwent coordinate transformation to extract a rectangular part containing a crack zone. Because the area of the crack zone was small and brightness was not uniform through the crack zone due to the uneven illumination over the zone, the extracted rectangular part was divided into blocks. Discrimination analysis was applied to the block unit to determine the threshold for binary-coding processing. Then each block was divided into sub- blocks and each sub-block underwent binary-coding processing. The binary images of sub-blocks thus obtained were corrected through a count filter and quadruple- connecting line-thinning processing based on the distances between adjacent feature points. The corrected images of sub-blocks were used to extract characteristics. The image used in this study was obtained by tracing the crack zone with white chalk.

In this study, characteristics were extracted based on the four criteria of the continuity, concentration, directionality (unidirectional or bi-directional), and types (hexagonal or linear) of cracks. As in the case of character patterns, a crack pattern of thin lines can be considered a set of directional linear elements and hence characteristics extraction by the projection histogram would be effective. Because the characteristics of projection distribution of a crack pattern reflect information on the positions and quantities of cracks, they can be used, as the characteristics quantities representing the continuity and concentration of cracks, for the purpose of classification of cracks. The characteristics quantities are the quantum numbers in accordance with the dimensionality of characteristics vectors.

The characteristics were extracted at feature points on boundary lines. Intersections, branch points, end points, and break points (points with a curvature over a certain value) were extracted from the crack pattern. Because the feature points of a crack pattern have information on the shapes of cracks, they can be used, as the characteristics quantities representing the directionality (unidirectional or bi- directional) and types (hexagonal or linear) of cracks, for the purpose of classification of cracks. The crack pattern was divided into blocks in accordance with the dimensionality of characteristics vectors and the frequencies of appearance of four types of feature points in each block were used as the quantities of characteristics.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £82 +P&P)