Keywords: crack detection, crack width measurement, deep learning, image processing, structural health monitoring, computer vision.

The presence of cracks in reinforced concrete structures is an indication of structural degradation. However, some cracks are more concerning than others, the maximum crack width is usually a good indication of the severity of a crack. The measurement of crack widths is typically done manually using crack width gauges, a method which tends to be tedious and susceptible to human error. Image processing algorithms have also been employed to measure crack widths, but these tend to give the measured crack widths in terms of number of pixels and not millimetres. It is difficult to assess the severity of the cracks when their widths are measured in pixel units. This is because design guides typically state the allowable crack widths are in units of millimetres rather than pixels. This paper presents a crack detection and quantification method that uses a deep learning model, a two-dimensional convolutional neural network, to detect the presence of cracks in captured images of reinforced concrete surfaces. The camera capturing the images has a laser pointer attached to it to project a circular laser light onto the measured plane. A relationship between the diameter of the laser projected on the measured plane and the distance to the measured plane was established. This relationship was used to convert the maximum pixel width, measured by using image processing algorithms in MATLAB, to millimetre width. The results of the study showed that the two-dimensional convolutional network was able to successfully detect cracks, with very high accuracy of 98.58%. The proposed method of converting pixel width to millimetre width also yielded positive results with percentage errors of less than 2%. Going beyond crack detection and measuring the crack widths in millimetres and not pixels can give a good insight into the condition of the structure in question, in accordance with international codes such as the Eurocodes. This simple, low-cost method was found to be very effective.

download the full-text of this paper (PDF, 7 pages, 386 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description